Nonlinear system voice noise reduction method and system based on maximum entropy criterion

Abstract

The invention provides a nonlinear system voice noise reduction method and system based on a maximum entropy criterion, and belongs to the technical field of voice communication active noise elimination, and the method comprises the steps: obtaining a noise source signal discrete value; performing Volterra series expansion on the signal discrete value to obtain a first input signal; decomposing the tap vector of the adaptive filter by using a Kronecker integral solution to obtain a plurality of decomposition filters; obtaining an error signal based on the first input signal and a decomposition filter; an error signal is collected and fed back to the filter, a tap weight vector of the decomposition filter at the next moment is obtained, and a maximum entropy criterion is adopted to update a weight vector of the adaptive filter; and n is updated to the next moment value until the noise reduction process is finished. According to the invention, the calculation complexity is reduced, and the signal-to-noise ratio of the noise reduction signal is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of active noise elimination in voice communication, and in particular relates to a method and system for voice noise reduction in a nonlinear system based on a maximum entropy criterion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In daily life and industrial application scenarios, there are generally environmental noises of various frequencies and amplitudes, which will inevitably mix with the voice signals collected by the microphone, resulting in a decrease in the purity of the voice signals. The interference of such environmental noise will have a significant negative impact on the clarity and intelligibility of the voice signals, which may cause the auditory system of the receiving listeners to bear excessive auditory load, causing hearing fatigue and even potential hearing damage.

[0004] To address this problem, the intervention of speech noise reduction technology is crucial. This technology aims to eliminate or greatly reduce the interference of background noise on speech signals through advanced signal processing algorithms. Its core goal is to separate and reconstruct high-fidelity "pure" speech signals from noisy speech signals. In this process, commonly used technical means include but are not limited to spectral subtraction, Wiener filtering, recursive least squares algorithm, and noise reduction models under the deep learning framework.

[0005] Speech noise reduction technology plays a vital role in voice communication systems. It significantly enhances the quality and clarity of voice communication by improving the signal-to-noise ratio of voice signals, and optimizes the interactive experience of both parties in communication. In addition, the effective application of this technology also helps to reduce the risks to auditory health caused by long-term exposure to noise environments, and ensures the reliability and comfort of voice communication. One of the typical traditional and widely used active noise cancellation methods is the filtered X least mean square (FXLMS) algorithm. However, the actual implementation of FXLMS is very unstable when dealing with main paths that show nonlinearity, secondary paths with nonlinear distortion, or reference noise generated by nonlinear processes.

[0006] The existing echo cancellation methods for resisting impact interference are as follows: [1] “Adaptive Volterra filters for active control of nonlinear noise processes”, (Tan, J. Jiang, IEEE Trans. Signal Process. 2001, 49 (8) 1667–1676) This paper appropriately expands the model and algorithm and solves the problem of dynamic nonlinearity. However, this paper only improves the FXLMS algorithm. Although it can handle the nonlinear main path, it has great disadvantages when dealing with the secondary path with nonlinear distortion or the reference noise generated by the nonlinear process.

[0007] [2] “Improving performance of FXLMS algorithm for active noise control of impulsive noise”, (MT Akhtar, W. Mitsuhashi, J. Sound Vib. 2009, 327 (3-5) 647–656), this paper proposes the VFXLMS algorithm to improve the robustness of the Volterra filter. The algorithm is based on the MSE criterion, but it is less effective in the impulsive noise environment; and the computational complexity of the Volterra filter increases exponentially with the length of the filter tap.

[0008] [3] “A family of adaptive volterra filters based on maximum correntropy criterion for improved active control of impulsive noise”, (G.Gowtham, S. Burra, A. Kar, J. Ostergaard, P. Sooraksa, V. Mladenovic, DBHaddad, Circuits Syst. Signal Process. 2022,41(2) 1019–1037), this paper proposes a variable step size VXRMC algorithm, which combines the RLS update method to solve the trade-off problem brought by the fixed learning rate, but still has the problem of high computational complexity. Summary of the invention

[0009] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes a nonlinear system speech denoising method and system based on the maximum entropy criterion, which improves the signal-to-noise ratio of active noise reduction of the nonlinear system and significantly reduces the computational complexity by utilizing a Volterra filter decomposed by tap coefficients and updating the decomposed filter based on the maximum entropy criterion.

[0010] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, a method for speech noise reduction in a nonlinear system based on a maximum entropy criterion is disclosed, comprising: Sampling the noise source, converting the analog signal of the noise source into an electrical signal to obtain a discrete value of the signal; The discrete value of the signal is expanded by Volterra series to obtain the first input signal; Decomposing the adaptive filter tap vector using Kronecker integral decomposition to obtain multiple decomposition filters; The first input signal is passed through a plurality of analysis filters to calculate a first output signal of the adaptive filter, the first output signal is amplified by a power amplifier, and then output to a main path via an active speaker; In the main path, an error signal is obtained based on the near-end signal and the first output signal, and a decomposition error signal of the decomposition filter is calculated; Collect the decomposition error signal, feed it back to the filter, obtain the tap weight vector of the decomposition filter at the next moment, and use the maximum entropy criterion to update the weight vector of the adaptive filter; Update n to the value at the next moment, and repeat the above steps until the denoising process is completed.

[0011] In a second aspect, a nonlinear system speech noise reduction system based on a maximum entropy criterion is disclosed, comprising: A signal acquisition module is configured to: sample the noise source through a reference microphone, convert the analog signal of the noise source into an electrical signal to obtain a discrete value of the signal; A signal expansion module is configured to: expand the discrete value of the signal through a Volterra series to obtain a first input signal; A filter decomposition module is configured to: decompose the adaptive filter tap vector using Kronecker integral decomposition to obtain a plurality of decomposition filters; A first calculation module: calculates a first output signal of an adaptive filter by passing a first input signal through a plurality of decomposition filters; the first output signal is amplified by a power amplifier and then output to a main path through an active speaker; A second calculation module: in the main path, obtaining an error signal based on the near-end signal and the first output signal, and calculating a decomposition error signal of the decomposition filter; The filter update module is configured to: collect the decomposition error signal through the error microphone, feed it back to the filter, obtain the tap weight vector of the decomposition filter at the next moment, and update the weight vector of the adaptive filter using the maximum entropy criterion; Update n to the value at the next moment, and repeat the above steps until the denoising process is completed.

[0012] In a third aspect, an electronic device is disclosed, including a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the above-mentioned nonlinear system speech denoising method based on the maximum entropy criterion are completed.

[0013] In a fourth aspect, a computer-readable storage medium is disclosed for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned nonlinear system speech denoising method based on the maximum entropy criterion are completed.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The invention provides a Volterra filter using tap coefficient decomposition, wherein the filter uses Kronecker integral decomposition to approximate a long-tap Volterra filter into a plurality of short-tap filters. By reducing the number of taps in the filter, the method greatly reduces the computational complexity of the Volterra filter.

[0015] For non-Gaussian noise sources in nonlinear active noise control, the decomposed filter is updated based on the maximum entropy criterion. The maximum entropy criterion is used to suppress large errors, thereby ignoring large error values ​​to ensure the stability of the filtering algorithm, thereby improving the signal-to-noise ratio.

[0016] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0018] Figure 1 This is a flow chart of the nonlinear system speech denoising method based on the maximum entropy criterion described in Embodiment 1 of the present invention.

[0019] Figure 2 This is a comparison chart of ANR curves of the speech noise reduction method described in Example 1 of the present invention and the prior art for nonlinear active noise reduction under a Logistic chaotic noise source.

[0020] Figure 3 This is a comparison chart of ANR curves of the speech noise reduction method described in Example 1 of the present invention and the prior art for nonlinear active noise reduction under a stable noise source.

[0021] Figure 4 This is a comparison chart of ANR curves of the speech noise reduction method described in Example 1 of the present invention and the prior art for nonlinear active noise reduction under a Gaussian noise source. DETAILED DESCRIPTION

[0022] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0023] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0024] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0025] Embodiment 1 In one or more embodiments, a method for reducing speech noise in a nonlinear system based on a maximum entropy criterion is disclosed, comprising the following steps: Step S1: sampling a noise source through a reference microphone, and converting an analog signal of the noise source into an electrical signal to obtain a discrete value of the signal.

[0026] Step S2: Expand the discrete value of the signal through the Volterra series to obtain a first input signal.

[0027] Step S201: Convert the current time n to The discrete value of the far-end signal at the moment constitutes the entire adaptive filter input signal at the current moment n , , where L represents the number of adaptive filter taps, and the superscript T represents the transpose. Represents the filter input length, n is the nth moment; Step S202: Input signal After Volterra series expansion, the first input signal is obtained , whose expression is:

[0028] Where P is the filter kernel order, is the filter input length.

[0029] The Volterra series can be used to accurately describe the nonlinear relationship between the input signal and the output signal. In active noise reduction, noise often has nonlinear characteristics. The Volterra series can better simulate this nonlinear relationship. By expanding it into a Volterra series, the linear and nonlinear components in the input signal can be analyzed to help understand the nonlinear response of the system. Before performing active noise reduction processing, the Volterra series is used to expand the input signal, and more effective noise reduction is achieved by better considering the characteristics of the signal.

[0030] Step S3: Use Kronecker integral decomposition to solve the adaptive filter tap vector Decompose to obtain multiple analysis filters, which are the first sub-filters with lengths of L1 and L2. and the second subfilter .

[0031] Adaptive filter tap vector for:

[0032] Where h is the filter weight coefficient; Using the Kronecker solution we get: ,in

[0033] Where k is the number of groups, which is a constant, and , where K is the array dimension, , and The dimensions are , The identity matrix of is the weight coefficient matrix obtained by the Klaunec product of the identity matrix and the first sub-filter; is the weight coefficient matrix obtained by the Klaunec product of the identity matrix and the second sub-filter; The meaning of is the k-group optimal solution of the first sub-filter, The meaning of is the k optimal solutions of the second sub-filter, and , ,in For the matrix The singular values ​​of and The matrices are The k-th column of the left singular value matrix and the right singular value matrix of .

[0034] Step S4: The first input signal By decomposition filter and Calculate the first output signal of the adaptive filter , the first output signal After the power is amplified by the power amplifier, it is output to the main path through the active speaker.

[0035] The calculation expression of the first output signal is:

[0036]

[0037] in, is a first output signal of the first sub-filter; is the first output signal of the second sub-filter; is the impulse response function; is the optimal solution for the first sub-filter tap vector; is the optimal solution for the second sub-filter tap vector; is the optimal solution of the first sub-input signal after decomposition; The meaning of is the optimal solution of the second sub-input signal after decomposition.

[0038] Step S5: In the main path, based on the near-end signal and the first output signal Get the error signal , and calculate the decomposition error signal of the decomposition filter with length L1 and L2 and .

[0039] Specifically: Near-end signal That is, the expected signal generated by the far-end signal (input signal) through the main path, and the near-end signal at the current time n Subtract the first output signal of the entire adaptive filter at the current time n , get the error signal at the current time n ; The error signal at the current time n Determine the decomposition error signal of the decomposition filter with length L1 and L2 as follows:

[0040]

[0041] The first decomposed error signal middle

[0042]

[0043]

[0044]

[0045] The second decomposed error signal middle

[0046]

[0047]

[0048] in, is the signal obtained by convolving the expanded input signal with the impulse response of the secondary path transfer function, and .

[0049] Step S6: The error microphone collects the decomposition error signal and feeds it back to the adaptive filter to obtain the tap weight vector of the decomposition filter at the next moment, and the weight vector of the adaptive filter is updated using the maximum entropy criterion.

[0050] Specifically, the MCC criterion cost function is selected and its gradient is calculated to obtain the weight update formula of the decomposition filter:

[0051]

[0052]

[0053] in, is the optimal solution for the first sub-filter tap vector at time n+1, is the optimal solution for the second sub-filter tap vector at time n+1, is the step size parameter, is the kernel width, It means the first sub-signal obtained by convolving the expanded input signal with the impulse response of the secondary path transfer function. It means the second sub-signal obtained by expanding the input signal and convolving it with the impulse response of the secondary path transfer function.

[0054] Correlation entropy is a method to measure the similarity between two random variables. It is insensitive to non-Gaussian noise, so the MCC algorithm can still maintain good performance in the presence of non-Gaussian noise. The calculation of correlation entropy involves a kernel function that measures the local similarity between two data points. This means that the MCC algorithm relies more on the local information of the data rather than the global statistical properties. Therefore, even if there are outliers or noise in the data, the MCC algorithm can still work effectively as long as the impact of these outliers is local.

[0055] Step S7: Update n to the value at the next moment, and repeat steps S1 to S6 until the noise reduction process is completed.

[0056] Specifically, the signs of the end of the noise reduction process include: (1) minimization of the error signal, (2) the weight coefficient of the adaptive filter converges to a stable value, (3) the system reaches a steady state, and (4) the noise reduction is terminated manually. When any of these signs is reached, the noise reduction process is terminated.

[0057] (1) Minimization of the error signal The judgment method is as follows: The active noise reduction system cancels out the noise by generating a sound with a phase opposite to the original noise. When the error signal of the system drops to a sufficiently low level, it indicates that the noise reduction effect has reached the best level, and the noise reduction process can be considered to be over. The error signal is the residual signal after the original noise is added to the anti-phase sound wave, expressed as:

[0058] Among them, α is the noise reduction end threshold. <α, that is, when the error signal is less than the threshold, the noise reduction process ends. The main goal of active noise reduction is to minimize the error signal. When the error signal reaches a very low level, it indicates that the noise has been effectively eliminated. The threshold α can be preset to any value according to the specific situation and is not limited in this embodiment.

[0059] (2) When the coefficients of the adaptive filter converge to a stable value, it indicates that the filter has found the optimal weight configuration to generate the anti-noise signal, and active noise reduction can be terminated at this time.

[0060] (3) When the system reaches a steady state, that is, when the statistical characteristics of the input signal no longer change, this indicates that the filter output has also stabilized, which indicates the end of active noise reduction.

[0061] (4) In practical applications, the end of active noise reduction may also depend on the user's feelings. If the user's artificial noise has been reduced to an acceptable level, the end of active noise reduction can be realized.

[0062] In this embodiment, a simulation experiment is carried out on the above method, and the noise reduction performance is compared with the VFXLMS algorithm, VFXLMP algorithm and VFXMCC algorithm in the prior art.

[0063] In simulation experiment =22, B1=23, B2=12, kernel width , select K=4.

[0064] Original disturbance for:

[0065]

[0066] P (n) represents the impulse response of the system:

[0067] Minimum phase quadratic path transfer function S(z):

[0068] The chaotic noise source is chosen as:

[0069] from Figure 2 It can be seen that under the Logistic chaotic noise source, the nonlinear system speech denoising method based on the maximum entropy criterion proposed in the present invention has the lowest steady-state average noise reduction (ANR), followed by VFXLMP, VFXMCC and VFXLMS algorithms. The method of the present invention has a significant steady-state advantage, especially reducing the steady-state ANR by about 5dB compared with the original algorithm.

[0070] from Figure 3 It can be seen that when the noise source is changed to the standard α-stable distribution with α=1.9, the VFXLMS and VFXLMP algorithms differ due to insufficient robustness, and the VFXMCC algorithm fluctuates. In contrast, the method proposed in the present invention has stability and minimal steady-state ANR.

[0071] from Figure 4 It can be seen that when Gaussian noise is input, the nonlinear system speech denoising method based on the maximum entropy criterion of the present invention shows the best denoising performance in the whole process, and the final ANR is about 5dB lower than VFXMCC, about 7dB lower than VFXLMP, and about 9dB lower than VFXLMS. The denoising performance of VFXMCC and VFXLMP algorithms is second-best, but still significantly better than VFXLMS. VFXLMS performs the worst, indicating that its denoising ability is limited when dealing with Gaussian noise sources.

[0072] Embodiment 2 In one or more embodiments, a nonlinear system speech denoising system based on a maximum entropy criterion is disclosed, specifically comprising: A signal acquisition module is configured to: sample the noise source through a reference microphone, convert the analog signal of the noise source into an electrical signal to obtain a discrete value of the signal; A signal expansion module is configured to: expand the discrete value of the signal through a Volterra series to obtain a first input signal; A filter decomposition module is configured to: decompose the adaptive filter tap vector using Kronecker integral decomposition to obtain a plurality of decomposition filters; A first calculation module: calculates a first output signal of an adaptive filter by passing a first input signal through a plurality of decomposition filters; the first output signal is amplified by a power amplifier and then output to a main path through an active speaker; A second calculation module: in the main path, obtaining an error signal based on the near-end signal and the first output signal, and calculating a decomposition error signal of the decomposition filter; The filter update module is configured to: collect the decomposition error signal through the error microphone, feed it back to the filter, obtain the tap weight vector of the decomposition filter at the next moment, and update the weight vector of the adaptive filter using the maximum entropy criterion; Update n to the value at the next moment, and repeat the above steps until the denoising process is completed.

[0073] Embodiment 3 This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps of the nonlinear system speech noise reduction method based on the maximum entropy criterion are completed.

[0074] Embodiment 4 This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned nonlinear system speech denoising method based on the maximum entropy criterion are completed.

[0075] 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.

[0076] 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 including 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.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide 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.

[0078] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for speech denoising in a nonlinear system based on the maximum entropy criterion, characterized in that: include: Sampling the noise source, converting the analog signal of the noise source into an electrical signal to obtain a discrete value of the signal; The discrete value of the signal is expanded by Volterra series to obtain the first input signal; Decomposing the adaptive filter tap vector using Kronecker integral decomposition to obtain multiple decomposition filters; The first input signal is passed through a plurality of analysis filters to calculate a first output signal of the adaptive filter, the first output signal is amplified by a power amplifier, and then output to a main path via an active speaker; In the main path, an error signal is obtained based on the near-end signal and the first output signal, and a decomposition error signal of the decomposition filter is calculated; Collect the decomposition error signal, feed it back to the filter, obtain the tap weight vector of the decomposition filter at the next moment, and use the maximum entropy criterion to update the weight vector of the adaptive filter; Update n to the value at the next moment, and repeat the above steps until the denoising process is completed.

2. A method for speech denoising in a nonlinear system based on the maximum entropy criterion as claimed in claim 1, characterized in that: The discrete value of the signal is expanded by Volterra series to obtain the first input signal, which is expressed as: in, is the first input signal; is the input signal; n for n moment; P is the filter kernel order; is the filter input length; T is the transpose.

3. A method for speech denoising in a nonlinear system based on the maximum entropy criterion as claimed in claim 1, characterized in that: The adaptive filter tap vector is decomposed using Kronecker integral decomposition to obtain multiple decomposition filters, which are expressed as follows: ,in in, is the adaptive filter tap vector; k is the number of groups, which is a constant, and , where K is the array dimension, ; and The dimensions are , The identity matrix of is the weight coefficient matrix obtained by the Klaunec product of the identity matrix and the first sub-filter; is the weight coefficient matrix obtained by the Klaunec product of the identity matrix and the second sub-filter; The meaning of is the k-group optimal solution of the first sub-filter, The meaning of is the k optimal solutions of the second sub-filter, and , ,in For the matrix The singular value of , h is the filter weight coefficient, and The matrices are The k-th column of the left singular value matrix and the right singular value matrix of .

4. A method for speech denoising in a nonlinear system based on the maximum entropy criterion as claimed in claim 1, characterized in that: The calculation expression of the first output signal is: in, is a first output signal of the first sub-filter; is the first output signal of the second sub-filter; is the impulse response function; is the optimal solution for the first sub-filter tap vector; is the optimal solution for the second sub-filter tap vector; is the optimal solution of the first sub-input signal after decomposition; The meaning of is the optimal solution of the second sub-input signal after decomposition.

5. The method for speech denoising in a nonlinear system based on the maximum entropy criterion according to claim 1, characterized in that: The error signal is calculated by subtracting the first output signal of the adaptive filter at the current time n from the expected signal at the current time n.

6. A method for speech denoising in a nonlinear system based on the maximum entropy criterion as claimed in claim 5, characterized in that: The decomposition error signal of the decomposition filter with length L1 and L2 is determined by the error signal at the current time n, and the formula is as follows: The first decomposed error signal middle The second decomposed error signal middle in, is the signal obtained by convolving the expanded input signal with the impulse response of the secondary path transfer function, and .

7. The method for speech denoising in a nonlinear system based on the maximum entropy criterion according to claim 1, characterized in that: The weight vector of the adaptive filter is updated by the maximum entropy criterion, and the expression is: in, is the optimal solution for the first sub-filter tap vector at time n+1, is the optimal solution for the second sub-filter tap vector at time n+1, is the step size parameter, is the kernel width, It means the first sub-signal obtained by convolving the expanded input signal with the impulse response of the secondary path transfer function. It means the second sub-signal obtained by expanding the input signal and convolving it with the impulse response of the secondary path transfer function.

8. A nonlinear system speech denoising system based on maximum entropy criterion, characterized in that: include: A signal acquisition module is configured to: sample the noise source through a reference microphone, convert the analog signal of the noise source into an electrical signal to obtain a discrete value of the signal; A signal expansion module is configured to: expand the discrete value of the signal through a Volterra series to obtain a first input signal; A filter decomposition module is configured to: decompose the adaptive filter tap vector using Kronecker integral decomposition to obtain a plurality of decomposition filters; A first calculation module: calculates a first output signal of an adaptive filter by passing a first input signal through a plurality of decomposition filters; the first output signal is amplified by a power amplifier and then output to a main path through an active speaker; A second calculation module: in the main path, obtaining an error signal based on the near-end signal and the first output signal, and calculating a decomposition error signal of the decomposition filter; The filter update module is configured to: collect the decomposition error signal through the error microphone, feed it back to the filter, obtain the tap weight vector of the decomposition filter at the next moment, and update the weight vector of the adaptive filter using the maximum entropy criterion; Update n to the value at the next moment, and repeat the above steps until the denoising process is completed.

9. An electronic device, characterized in that: The invention comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the nonlinear system speech denoising method based on the maximum entropy criterion as described in any one of claims 1 to 7 is completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the nonlinear system speech denoising method based on the maximum entropy criterion as described in any one of claims 1 to 7.

Images