Voice noise reduction system and method based on spectral subtraction noise reduction parameter optimization

Abstract

The invention relates to a voice noise reduction system and method based on spectral subtraction noise reduction parameter optimization, an optimization device is arranged, noise reduction parameters introduced by spectral subtraction are optimized by using a covariance matrix self-adaptive evolutionary strategy algorithm, parameter solidification is avoided, a noise reduction device is further arranged, the noise reduction parameters obtained by the optimization device are substituted into spectral subtraction, and the noise reduction parameters are optimized by using the covariance matrix self-adaptive evolutionary strategy algorithm. And noise reduction is carried out on the input audio through spectral subtraction so as to output the audio signal after noise reduction, so that the parameter combination of the spectral subtraction can be optimized, the optimal condition suitable for audio noise reduction is found, and the adaptability of a non-stationary noise scene is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of audio signal processing, and more particularly, to a voice noise reduction system and method based on optimized noise reduction parameters of spectral subtraction. Background Art

[0002] Voice noise reduction technology is a key technology for eliminating noise in audio signal propagation. This technology is widely used in the field of enhancing target voice signals in noisy environments. Spectral Subtraction is a classic method of this technology.

[0003] The basic idea of spectral subtraction is to utilize the statistical difference between noise and voice signals in the frequency domain, subtract the estimated noise power spectrum from the short-time power spectrum of the voice signal, so as to obtain a relatively pure voice signal. In this process, noise reduction parameters need to be introduced to complete the estimation of the noise power spectrum.

[0004] However, traditional spectral subtraction often uses an empirically preset parameter combination for the introduced noise reduction parameters for noise reduction processing. Although it can ensure the basic noise reduction effect, the parameter solidification leads to extremely poor adaptability to sudden noise and non-stationary noise scenarios, and the improvement amplitude of the signal-to-noise ratio is limited. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to overcome the technical defects in the existing spectral subtraction, which are poor adaptability to sudden noise and non-stationary noise scenarios due to parameter solidification, and limited improvement amplitude of the signal-to-noise ratio. To overcome the above defects of the prior art, the present invention provides a voice noise reduction system and method based on optimized noise reduction parameters of spectral subtraction, including a voice noise reduction system based on optimized noise reduction parameters of spectral subtraction and a voice noise reduction method based on optimized noise reduction parameters of spectral subtraction.

[0006] A voice noise reduction system based on optimized noise reduction parameters of spectral subtraction provided by the present invention includes:

[0007] An optimization device, configured to obtain noise reduction parameters by using a covariance matrix adaptation evolution strategy algorithm with a test audio file;

[0008] A noise reduction device, electrically connected to the optimization device, configured to use the noise reduction parameters obtained by the optimization device as the noise reduction parameters of spectral subtraction, and perform noise reduction on the input audio through the spectral subtraction to output a noise-reduced audio signal.

[0009] The voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters disclosed by the present invention sets an optimization device to optimize the noise reduction parameters introduced by spectral subtraction by using the Covariance Matrix Adaptation Evolution Strategy algorithm (CMAES), avoiding parameter solidification. A noise reduction device is also set to substitute the noise reduction parameters obtained by the optimization device into spectral subtraction, and perform noise reduction on the input audio through spectral subtraction to output a noise-reduced audio signal. Therefore, it can optimize the parameter combination of spectral subtraction, find the best situation suitable for audio noise reduction, improve the adaptability to non-stationary noise scenarios, further improve the noise reduction effect of spectral subtraction, and make the output noise-reduced audio signal have the effect of low noise interference, thereby overcoming the technical defects existing in spectral subtraction in the prior art, such as poor adaptability to sudden noise and non-stationary noise scenarios due to parameter solidification, and limited improvement in signal-to-noise ratio.

[0010] In a possible implementation manner, the noise reduction parameters include a parameter over-subtraction factor, a minimum value limit factor, a smoothing factor, a noise estimation smoothing coefficient, and a noise estimation number of frames; these parameters cover the introduced parameters of spectral subtraction, so as to ensure the optimization of all introduced parameters in spectral subtraction.

[0011] In a possible implementation manner, the optimization device is set to perform the following steps:

[0012] A1: Use the signal-to-noise ratio difference between the noise-reduced audio signal and the audio signal before noise reduction as the objective function, and set the previous mean value, the previous step size, the previous path, the loop limit condition, the previous covariance matrix evolution path, and the previous covariance matrix matching the number of the noise reduction parameters.

[0013] A2: Generate multiple candidate solutions in the domain of the multivariate normal distribution function determined by the previous mean value, the previous step size, and the previous covariance matrix, and sort these candidate solutions in ascending order according to the objective function values.

[0014] A3: Determine whether the loop limit condition is satisfied.

[0015] If so, take the candidate solution with the smallest objective function value as the noise reduction parameter.

[0016] If not, execute the next step.

[0017] A4: Select several candidate solutions with higher rankings from these candidate solutions, and use these candidate solutions to obtain the current mean value through the maximum likelihood estimation method.

[0018] A5: Update the current path by using the step size evolution path update formula, the previous path, the current mean value, and the previous mean value.

[0019] A6: Obtain the current step size through the step size update formula, using the current path and the previous step size.

[0020] A7: Obtain the current covariance matrix evolution path through the covariance matrix evolution path update formula, using the previous covariance matrix evolution path, the previous mean, the current mean, and the previous step size.

[0021] A8: Obtain the current covariance matrix through the covariance matrix construction formula, using the current covariance matrix evolution path, the previous covariance matrix, the current path, the current mean, and the previous mean.

[0022] A9: Take the current covariance matrix as the previous covariance matrix, take the current covariance matrix evolution path as the previous covariance matrix evolution path, take the current step size as the previous step size, take the current path as the previous path, take the current mean as the previous mean, and loop back to execute the step A2.

[0023] This solution can avoid parameter solidification and obtain noise reduction parameters. Find the best case suitable for audio noise reduction.

[0024] In a possible implementation, in the step A4, the calculation formula for obtaining the current mean through the maximum likelihood estimation method is as follows:

[0025]

[0026] And In the formula,

[0027] g represents the number of iterations;

[0028] λ represents the index for generating candidate solutions in the step A2;

[0029] μ represents the number of candidate solutions selected with a higher ranking in the step A4;

[0030] m represents the mean;

[0031] ω i represents the weight;

[0032] x i；λ represents the i-th candidate solution selected with a higher ranking in the step A4.

[0033] In a possible implementation, the calculation formula for the step size evolution path update formula is as follows:

[0034]

[0035] In the formula,

[0036] Pσ Represents a path;

[0037] c σ Represents a path update factor;

[0038] σ represents the step size;

[0039] C represents the covariance matrix.

[0040] In a possible implementation, the arithmetic formula of the step size update formula is as follows:

[0041]

[0042] In the formula,

[0043] c σ Represents the step size update factor;

[0044] d σ Represents the damping coefficient;

[0045] E∥N(0,I)∥ represents the expected length of the normal distribution random vector norm.

[0046] In a possible implementation, the calculation formula of the covariance matrix evolution path update formula is as follows:

[0047]

[0048] In the formula,

[0049] P c Represents the covariance matrix evolution path;

[0050] c c Represents the covariance matrix evolution path update factor.

[0051] In a possible implementation, the calculation formula of the covariance matrix construction formula is as follows:

[0052]

[0053] In the formula,

[0054] c cov Represents the update factor of the covariance matrix;

[0055] μ cov Represents the adaptive factor of the covariance matrix.

[0056] In a possible implementation, the noise reduction device is set to perform the following steps:

[0057] B1 performs frame splitting, windowing, and short-time Fourier transform on the input audio in sequence to obtain the signal in the frequency domain of the input audio;

[0058] B2: Extract the number of noise estimation frames from the noise reduction parameters obtained from the optimization device, and then use the audio of the first noise estimation frames at the beginning of the input audio as the non-speech segment to obtain the initial noise power spectrum;

[0059] B3: Extract the smoothing factor and the noise estimation smoothing coefficient from the noise reduction parameters obtained from the optimization device, and based on the initial noise power spectrum, perform noise smoothing estimation through the recursive smoothing update formula to obtain the noise power spectrum;

[0060] B4: Perform an inverse Fourier transform on the noise power spectrum obtained in step B3 to obtain the noise signal in the time domain;

[0061] B5: Extract the parameter over-subtraction factor and the minimum value limit factor from the noise reduction parameters obtained from the optimization device, and use the noise signal in the time domain to obtain the gain function by the spectral subtraction formula;

[0062] B6: Use the gain function to perform amplitude modulation on the signal in the frequency domain obtained in step A1 to obtain the noise-reduced audio spectrum;

[0063] B7: Perform an inverse Fourier transform on the noise-reduced audio spectrum to output the noise-reduced audio signal.

[0064] Another technical solution of the present invention is to provide a speech noise reduction method based on the optimization of spectral subtraction noise reduction parameters, including the following steps:

[0065] S1: Use the covariance matrix adaptive evolution strategy algorithm through an optimization device to obtain noise reduction parameters based on the test audio file;

[0066] S2: Use the noise reduction parameters obtained in step S1 as the noise reduction parameters of the spectral subtraction method through a noise reduction device, and perform noise reduction on the input audio through the spectral subtraction method to obtain the noise-reduced audio signal.

[0067] For the method disclosed in the present application, by performing step S1 to optimize the noise reduction parameters introduced by the spectral subtraction method using the covariance matrix adaptive evolution strategy algorithm, and then through step S2 to substitute the noise reduction parameters obtained by the optimization device into the spectral subtraction method, and perform noise reduction on the input audio through the spectral subtraction method to output the noise-reduced audio signal, it is possible to optimize the parameter combination of the spectral subtraction method, find the best case suitable for audio noise reduction, further improve the noise reduction effect of the spectral subtraction method, and make the output noise-reduced audio signal have the effect of low noise interference. Description of the Drawings

[0068] Figure 1 It is a schematic structural diagram of a speech noise reduction system based on the optimization of spectral subtraction noise reduction parameters disclosed in the embodiments of the present application;

[0069] Figure 2It is the flowchart of the optimization device operation disclosed in the embodiments of the present application;

[0070] Figure 3 It is the flowchart of the noise reduction device operation disclosed in the embodiments of the present application. Specific embodiments

[0071] First of all, those skilled in the art should understand that these embodiments are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can adjust it according to needs to adapt to specific application scenarios.

[0072] In the embodiments of the present application, unless otherwise clearly specified and limited, the electrical connection between the first feature and the second feature means that there is an electrical signal transmission between the first feature and the second feature, that is, there is an electrical relationship, and the way to realize the electrical signal transmission can be wire electrical connection, radio connection, electrical connection of electromagnetic media (such as semiconductors), communication realized by channels, etc.

[0073] In the embodiments of the present application, unless otherwise clearly specified and limited, the first feature being "on" or "under" the second feature can be that the first and second features are in direct contact, or the first and second features are indirectly in contact through an intermediate medium. Moreover, the first feature being "above", "over" and "on" the second feature can be that the first feature is directly above or obliquely above the second feature, or only indicates that the first feature is at a higher horizontal height than the second feature. The first feature being "under", "beneath" and "under" the second feature can be that the first feature is directly below or obliquely below the second feature, or only indicates that the first feature is at a lower horizontal height than the second feature.

[0074] The present application will be further described in detail below with reference to the drawings and specific embodiments.

[0075] See Figures 1 to 3 , the embodiments of the present application disclose a voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters. The structural schematic diagram of the voice noise reduction system is as Figure 1 shown. The voice noise reduction system includes an optimization device and a noise reduction device, and the noise reduction device is electrically connected to the optimization device.

[0076] In this voice noise reduction system, the optimization device is configured to obtain noise reduction parameters by using the covariance matrix adaptive evolution strategy algorithm with the test audio file. In this embodiment, the noise reduction parameters include a parameter over-subtraction factor, a minimum value limit factor, a smoothing factor, a noise estimation smoothing coefficient, and a noise estimation number of frames.

[0077] See Figure 2 , in this embodiment, the optimization device is configured to perform the following steps:

[0078] A1: Use the signal-to-noise ratio difference between the denoised audio signal and the audio signal before denoising as the objective function, and set the previous mean, the previous step size, the previous path, the loop limit condition, the previous covariance matrix evolution path, and the previous covariance matrix that matches the number of denoising parameters.

[0079] A2: Generate multiple candidate solutions in the domain of the multivariate normal distribution function determined by the previous mean, the previous step size, and the previous covariance matrix, and sort these candidate solutions in ascending order according to the objective function value. The expression of the multivariate normal distribution is as follows:

[0080]

[0081] In the formula,

[0082] ~ represents following the same distribution;

[0083] g represents the number of iterations;

[0084] B represents an orthogonal matrix; the columns of matrix B are the unit eigenvectors of the covariance matrix C;

[0085] D represents a diagonal matrix, and its diagonal elements are the square roots of the eigenvalues of matrix C;

[0086] I represents the identity matrix.

[0087] A3: Determine whether the loop limit condition is satisfied,

[0088] If so, take the candidate solution with the smallest objective function value as the denoising parameter;

[0089] If not, execute the next step.

[0090] In this embodiment, the loop limit condition includes:

[0091] Check the step size convergence condition: σ (g) <∈1 (whether the step size is less than the threshold ∈1);

[0092] Check the fitness change condition: f k <∈2 (whether the fitness change is less than the threshold ∈2);

[0093] Check the maximum iteration number condition: g≥g max (whether the maximum iteration number g max ) is reached;

[0094] If none of the above three limit conditions are met, it is considered that the loop limit condition is not satisfied. If one of them is satisfied, exit the loop to end the optimization. Output the denoising parameter.

[0095] A4: Select several candidate solutions with higher rankings from these candidate solutions, and use these several candidate solutions to obtain the current mean value through the maximum likelihood estimation method.

[0096] In step A4, the calculation formula for obtaining the current mean value through the maximum likelihood estimation method is as follows:

[0097]

[0098] And In the formula,

[0099] λ represents the index for generating candidate solutions in step A2;

[0100] μ represents the number of candidate solutions with higher rankings selected in step A4;

[0101] m represents the mean value;

[0102] ω i represents the weight;

[0103] x i；λ represents the i-th candidate solution with higher rankings selected in step A4.

[0104] A5: Update the current path using the step-size evolution path update formula, and obtain the current path by using the previous path, the current mean value, and the previous mean value; the calculation formula of the step-size evolution path update formula is as follows:

[0105]

[0106] In the formula,

[0107] P σ represents the path;

[0108] c σ represents the path update factor;

[0109] σ represents the step size;

[0110] C represents the covariance matrix.

[0111] A6: Update the current step size using the step-size update formula, and obtain the current step size by using the current path and the previous step size; the calculation formula of the step-size update formula is as follows:

[0112]

[0113] In the formula,

[0114] c σ represents the step-size update factor;

[0115] d σ represents the damping coefficient;

[0116] $E\|\mathbf{N}(0, \mathbf{I})\|$ represents the expected length of the norm of a normal distribution random vector.

[0117] A7: Through the covariance matrix evolution path update formula, the current covariance matrix evolution path is obtained using the previous covariance matrix evolution path, the previous mean, the current mean, and the previous step size; the calculation formula of the covariance matrix evolution path update formula is as follows:

[0118]

[0119] In the formula,

[0120] $\mathbf{P}$ c represents the covariance matrix evolution path;

[0121] $c$ c represents the covariance matrix evolution path update factor.

[0122] A8: Through the covariance matrix construction formula, the current covariance matrix is obtained using the current covariance matrix evolution path, the previous covariance matrix, the current path, the current mean, and the previous mean; the calculation formula of the covariance matrix construction formula is as follows:

[0123]

[0124] In the formula,

[0125] $c$ cov represents the update factor of the covariance matrix;

[0126] $\mu$ cov represents the adaptive factor of the covariance matrix.

[0127] A9: Take the current covariance matrix as the previous covariance matrix, the current covariance matrix evolution path as the previous covariance matrix evolution path, the current step size as the previous step size, the current path as the previous path, the current mean as the previous mean, and then loop back to execute step A2.

[0128] In this voice noise reduction system, the noise reduction device is set to use the noise reduction parameters obtained by the optimization device as the noise reduction parameters of the spectral subtraction method, and perform noise reduction on the input audio through the spectral subtraction method to output the noise-reduced audio signal.

[0129] See Figure 3 , in this embodiment, the noise reduction device is set to perform the following steps:

[0130] B1 Frame, window, and perform short-time Fourier transform on the input audio in sequence to obtain the signal in the frequency domain of the input audio.

[0131] Specifically, first frame and window the input audio x(t) of the noisy speech signal, and then calculate its short-time Fourier transform (STFT):

[0132] X(f,t) = F{x(t)},

[0133] where: X(f,t) is the representation of the noisy speech in the frequency domain. f is the frequency index, and t is the time index.

[0134] B2: Extract the number of noise estimation frames from the noise reduction parameters obtained by the optimization device, and then use the audio of the first number of noise estimation frames at the beginning of the input audio as the non-speech segment to obtain the initial noise power spectrum.

[0135] Let n represent the number of noise estimation frames. The first n frames at the start of the speech are usually non-speech segments. The initial noise power spectrum can be obtained through the following calculation formula:

[0136]

[0137] represents the noise power spectrum.

[0138] B3: Extract the smoothing factor and the noise estimation smoothing coefficient from the noise reduction parameters obtained by the optimization device, and based on the initial noise power spectrum, perform noise smoothing estimation through the recursive smoothing update formula to obtain the noise power spectrum.

[0139] Specifically, when the noise changes over time, the noise power spectrum is obtained through the following recursive smoothing update formula:

[0140]

[0141] In the formula, γ represents the noise estimation smoothing factor, with a range of 0.9 - 0.99, which controls the smoothing degree of the noise estimation. The larger the value, the stronger the smoothing effect, but it may lag behind the noise change. The number of noise estimation frames affects the estimation stability, and a larger number of noise estimation frames provides a more stable noise estimation. The noise estimation smoothing coefficient k can be used to adjust the estimated value.

[0142] B4: Perform an inverse Fourier transform on the noise power spectrum obtained in step B3 to obtain the noise signal in the time domain.

[0143] Specifically, the noise power spectrum can be transformed into the noise signal n(t) in the time domain through the inverse Fourier transform,

[0144]

[0145] B5: Extract the parameter subtraction factor and the minimum value limit factor from the noise reduction parameters obtained by the optimization device, and use the noise signal in the time domain to obtain the gain function with the spectral subtraction formula.

[0146] The calculation of the gain function is mainly affected by the over-subtraction factor α and the spectral lower limit threshold β. The gain function is calculated according to the spectral subtraction formula:

[0147]

[0148] Where: α (over-subtraction factor) controls the degree of noise reduction. Usually, 1 ≤ α ≤ 2. The larger the value, the stronger the noise reduction, but it may cause speech distortion. β (noise retention factor) prevents the gain from being too low and avoids speech distortion. Usually, β = 0.002 - 0.01.

[0149] B6: Use the gain function to amplitude-modulate the signal in the frequency domain obtained in step A1 to obtain the noise-reduced audio spectrum.

[0150] Specifically, use the gain function to adjust the spectrum of the noisy audio:

[0151] Y(f,t) = G(f,t) · X(f,t),

[0152] In the formula: Y(f,t) represents the noise-reduced audio spectrum.

[0153] B7: Perform an inverse Fourier transform on the noise-reduced audio spectrum to output the noise-reduced audio signal. That is, finally, obtain the noise-reduced speech signal y(t) through the inverse Fourier transform,

[0154] y(t) = F -1 {Y(f,t)}.

[0155] Generally, the difference in the signal-to-noise ratio (SNR) before and after noise reduction is used to represent the noise reduction effect of the spectral subtraction method, that is, the objective function of step A1. The signal-to-noise ratio before noise reduction is:

[0156]

[0157] The signal-to-noise ratio after noise reduction is:

[0158]

[0159] Where e(t) = x(t) - y(t).

[0160] Correspondingly, the objective function of this embodiment is:

[0161]

[0162] The following will further disclose the usage method of the voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters in this embodiment. The method includes the following steps: S1: Obtain the noise reduction parameters based on the test audio file by using the covariance matrix adaptive evolution strategy algorithm through the optimization device; S2: Use the noise reduction parameters obtained in step S1 as the noise reduction parameters of the spectral subtraction method through the noise reduction device, and perform noise reduction on the input audio through the spectral subtraction method to obtain the noise-reduced audio signal.

[0163] In the voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters disclosed in this embodiment, by setting an optimization device, the covariance matrix adaptive evolution strategy algorithm is used to optimize the noise reduction parameters introduced by the spectral subtraction method, avoiding parameter solidification. A noise reduction device is also set, and the noise reduction parameters obtained by the optimization device are brought into the spectral subtraction method, and the input audio is noise-reduced through the spectral subtraction method to output the noise-reduced audio signal. Therefore, the parameter combination of the spectral subtraction method can be optimized, the best situation suitable for audio noise reduction can be found, the adaptability to non-stationary noise scenarios can be improved, the noise reduction effect of the spectral subtraction method can be further improved, and the output noise-reduced audio signal has the effect of low noise interference, thereby overcoming the technical defects of the spectral subtraction method in the prior art, such as poor adaptability to sudden noise and non-stationary noise scenarios due to parameter solidification, and limited improvement in signal-to-noise ratio.

[0164] In the description of the embodiments of this application, it should be noted that in the description of this application, terms such as "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is only for convenience of description and does not indicate or imply that the device or component must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of this application.

[0165] In the description of this application, the descriptions with reference to terms such as "one embodiment", "some embodiments", "in this embodiment", "specific examples", or "some examples" mean that the specific features, mechanisms, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0166] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A voice noise reduction system based on optimization of spectral subtraction noise reduction parameters, characterized in that, Including: An optimization device, configured to obtain noise reduction parameters by using a covariance matrix adaptation evolution strategy algorithm with a trial audio file; A noise reduction device, electrically connected to the optimization device, configured to use the noise reduction parameters obtained by the optimization device as the noise reduction parameters of spectral subtraction, and perform noise reduction on the input audio by the spectral subtraction to output a noise-reduced audio signal.

2. The voice noise reduction system based on spectral subtraction noise reduction parameter optimization according to claim 1, wherein, The noise reduction parameters include a parameter over-subtraction factor, a minimum value limit factor, a smoothing factor, a noise estimation smoothing coefficient, and a noise estimation frame number.

3. The voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters according to claim 2, characterized in that, The optimization device is configured to perform the following steps: A1: Use the signal-to-noise ratio difference between the noise-reduced audio signal and the audio signal before noise reduction as the objective function, and set the previous mean value, the previous step size, the previous path, the loop limit condition, the previous covariance matrix evolution path, and the previous covariance matrix that matches the number of the noise reduction parameters; A2: Generate a plurality of candidate solutions in the domain of the multivariate normal distribution function determined by the previous mean value, the previous step size, and the previous covariance matrix, and sort these candidate solutions in ascending order according to the objective function value; A3: Determine whether the loop limit condition is satisfied, If so, use the candidate solution with the smallest objective function value as the noise reduction parameters; If not, perform the next step; A4: Select a plurality of candidate solutions with a higher ranking from these candidate solutions, and use these candidate solutions to obtain the current mean value by the maximum likelihood estimation method; A5: Use the step size evolution path update formula to obtain the current path by using the previous path, the current mean value, and the previous mean value; A6: Use the step size update formula to obtain the current step size by using the current path and the previous step size; A7: Use the covariance matrix evolution path update formula to obtain the current covariance matrix evolution path by using the previous covariance matrix evolution path, the previous mean value, the current mean value, and the previous step size; A8: Use the covariance matrix construction formula to obtain the current covariance matrix by using the current covariance matrix evolution path, the previous covariance matrix, the current path, the current mean value, and the previous mean value; A9: Use the current covariance matrix as the previous covariance matrix, use the current covariance matrix evolution path as the previous covariance matrix evolution path, use the current step size as the previous step size, use the current path as the previous path, use the current mean value as the previous mean value, and return to execute the step A2.

4. The voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters according to claim 3, wherein, In the step A4, the calculation formula for obtaining the current mean value by the maximum likelihood estimation method is as follows: and ω i > 0, In the formula, g represents the number of iterations; λ represents the index for generating candidate solutions in the step A2; μ represents the number of candidate solutions with a higher ranking selected in the step A4; m represents the mean value; ω i represents a weight value; x i；λ represents the i-th candidate solution with a relatively high ranking selected in the step A4.

5. The voice noise reduction system based on optimized spectral subtraction noise reduction parameters according to claim 4, characterized in that The calculation formula of the step size evolution path update formula is as follows: In the formula, P σ Represents a path; c σ represents a path update factor; σ represents the step size; C represents the covariance matrix.

6. The voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters according to claim 5, characterized in that, The calculation formula of the step size update formula is as follows: In the formula, c σ represents the step size update factor; d σ represents the damping coefficient; E||N(0,I)|| represents the expected length of the normal distribution random vector norm.

7. The voice noise reduction system based on optimized spectral subtraction noise reduction parameters according to claim 6, characterized in that The calculation formula of the covariance matrix evolution path update formula is as follows: In the formula, P c Represents the evolutionary path of the covariance matrix; c c Represents the covariance matrix evolution path update factor.

8. The voice noise reduction system based on optimized spectral subtraction noise reduction parameters according to claim 7, characterized in that, The calculation formula of the covariance matrix construction formula is as follows: In the formula, c cov Represents the update factor of the covariance matrix; μ cov Represents the adaptive factor of the covariance matrix.

9. The voice noise reduction system based on the optimization of spectral subtraction noise reduction parameters according to claim 8, characterized in that, The noise reduction device is set to perform the following steps: B1: Frame, window, and perform short-time Fourier transform on the input audio in sequence to obtain the signal in the frequency domain of the input audio; B2: Extract the number of noise estimation frames from the noise reduction parameters obtained by the optimization device, and then use the audio of the first noise estimation frames at the beginning of the input audio as the non-speech segment to obtain the initial noise power spectrum; B3: Extract the smoothing factor and the noise estimation smoothing coefficient from the noise reduction parameters obtained by the optimization device, and based on the initial noise power spectrum, perform noise smoothing estimation through the recursive smoothing update formula to obtain the noise power spectrum; B4: Perform inverse Fourier transform on the noise power spectrum obtained in step B3 to obtain the noise signal in the time domain; B5: Extract the parameter subtraction factor and the minimum value limit factor from the noise reduction parameters obtained by the optimization device, and use the noise signal in the time domain to obtain the gain function with the spectral subtraction formula; B6: Use the gain function to perform amplitude modulation on the signal in the frequency domain obtained in step A1 to obtain the noise-reduced audio spectrum; B7: Perform inverse Fourier transform on the noise-reduced audio spectrum to output the noise-reduced audio signal.

10. A voice noise reduction method based on optimization of spectral subtraction noise reduction parameters, characterized in that, Applicable to the speech noise reduction system based on spectral subtraction noise reduction parameter optimization described in any one of claims 1-9, including the following steps: S1: Obtain the noise reduction parameters through the optimization device using the covariance matrix adaptive evolution strategy algorithm based on the test audio file; S2: Use the noise reduction parameters obtained in step S1 as the noise reduction parameters of the spectral subtraction method through the noise reduction device, and perform noise reduction on the input audio through the spectral subtraction method to obtain the noise-reduced audio signal.

Images