Echo cancellation and noise reduction method and system for steel mill workshop environment sound

Abstract

The invention belongs to the field of industrial noise control, and relates to an echo cancellation and noise reduction method and system for environmental sound of a steel mill workshop. The method comprises the following steps: (1) acquiring a sound signal of an on-site monitoring point in a steel mill workshop environment in a working process in real time; the field monitoring point comprises each noise source and each target source; (2) aligning the original sound signal and the workshop echo signal from the same on-site monitoring point in time to obtain an aligned signal; (3) filtering the aligned signal obtained in the step (2) so as to suppress noise and echo signals superposed in a channel; and (4) training a neural network mask, processing the sound signal by using a mask frequency domain to suppress a residual noise signal, and outputting a clean sound signal. According to the method provided by the invention, the application limitation of a traditional sound noise reduction technology in a complex industrial environment is effectively solved.

Description

Technical Field

[0001] The invention belongs to the field of industrial noise control, and in particular relates to an echo cancellation and noise reduction method and system for environmental sound in a steel plant workshop. Background Art

[0002] In heavy industrial fields such as steel manufacturing, the ambient noise level is usually very high, which not only affects the health and safety of workers, but also seriously interferes with the early detection of equipment failures and the effective operation of sound monitoring systems. Traditional sound noise reduction technologies such as active noise control and passive acoustic isolation focus on reducing the impact of noise on the human ear, but are not enough to handle the extraction and enhancement of signals in complex industrial noise environments.

[0003] In heavy industrial environments such as steel mills, there are multiple noise sources, including mechanical operation, impact and electrical noise, which are characterized by a wide frequency range, high intensity and rapid changes. In addition, the sound signals in these environments usually contain important operational information, such as equipment operating status and fault warnings. Therefore, effective technology is needed to improve the recognizability and clarity of these signals. At the same time, the workshop environment is prone to echo and reverberation, and the enhancement of acoustic signal quality is a difficult problem that needs to be solved urgently.

[0004] Currently, most sound noise reduction technologies focus on simple linear processing methods, such as filtering and echo cancellation, which are effective in dealing with single or predictive noise sources, but have limited effects in multiple sound sources and high-dynamic environments. For example, linear filters cannot adapt to the rapid changes in environmental noise, and although traditional neural network methods can improve the ability to process complex signals, they still face challenges in computational efficiency and resource consumption in real-time applications.

[0005] In addition, existing technologies often fail to fully utilize the frequency and time correlation of sound signals, resulting in the inability to effectively extract useful signals from high-noise backgrounds. Therefore, developing a technology that can effectively process noise in the complex acoustic environment of a steel plant is of great significance for improving production safety and efficiency. Summary of the invention

[0006] In response to the above technical problems, the present invention discloses an echo cancellation and noise reduction method and system for the environmental sound of a steel plant workshop; by integrating a delay compensation module, a linear filter module and a residual noise suppression module based on a deep neural network, it effectively solves the limitations of the application of traditional sound noise reduction technology in complex industrial environments.

[0007] In order to achieve the above technical objectives, the present invention provides the following technical solutions:

[0008] A method for echo cancellation and noise reduction of environmental sound in a steel plant workshop, the method comprising the following steps:

[0009] (1) Real-time acquisition of sound signals of on-site monitoring points in a steel plant workshop environment during operation; the on-site monitoring points include various noise sources and various target sources;

[0010] (2) Aligning the original acoustic signal and the workshop echo signal from the same on-site monitoring point in time to obtain an aligned signal;

[0011] (3) filtering the aligned signal obtained in step (2) to suppress the noise and echo signals superimposed in the channel;

[0012] (4) training the neural network mask, filtering the sound signal processed by the mask frequency domain processing step (3) to suppress the residual noise signal, and outputting a clean sound signal.

[0013] Furthermore, in step (2), a generalized cross-correlation and phase transformation algorithm is used to calculate the time delay of the original sound signal and the reverberation of the workshop echo signal generated by the on-site workshop.

[0014] Furthermore, step (2) specifically includes:

[0015] (2.1) Define generalized cross-correlation, calculate the statistical correlation between the original sound signal of the steel plant operation workshop and the echo signal of the workshop, and establish the association between the signals; at each time step, update the cross-correlation value between the original sound signal of the steel plant operation workshop and the echo signal of the workshop;

[0016] (2.2) The frequency domain signal is converted back to the time domain signal through inverse fast Fourier transform; in the time domain signal, the maximum position of the cross-correlation value is found, the delay time between the signals is calculated, and the workshop echo signal is adjusted so that the workshop echo signal is aligned in time with the corresponding original sound signal.

[0017] Furthermore, step (2.2) specifically includes:

[0018] Perform fast Fourier transform on the original sound signal x1(t) and the workshop echo signal x2(t) from the same on-site monitoring point to obtain X1(f) and X2(f);

[0019] Cross power spectral density of two signals The calculation formula is:

[0020]

[0021] in, is the conjugate of X2(f);

[0022] Apply phase transformation to obtain the regularized cross power spectral density

[0023]

[0024] The regularized cross power spectral density Perform an inverse fast Fourier transform to convert from the frequency domain to the time domain, obtaining the cross-correlation function

[0025]

[0026] Find the location of the peak in the cross-correlation function τ max , position τ max That is, the estimated delay between the original sound signal x1(t) and the workshop echo signal x2(t);

[0027] According to the calculated delay τ max , adjust the time axis of the signal x2(t) so that x1(t) and x2(t+τ max ) are aligned in time to obtain the aligned signal x(t).

[0028] Furthermore, in step (3), a weighted recursive least squares filter is used for filtering.

[0029] Furthermore, step (3) specifically includes:

[0030] Take the aligned signal x(t) as input, initialize the filter weight vector w(0) and the inverse of the correlation matrix P(0); usually P(0) is set to a large scalar multiplied by the unit matrix, such as δI, where δ is a large positive number; specifically, δ can be set to 10^6 to enhance the adaptability of the algorithm, which is suitable for situations where the input signal has large dynamic changes and requires faster convergence.

[0031] Calculate the gain vector k(n):

[0032]

[0033] Among them, the input vector of each time step n is x(n), x(n) means the alignment signal of all on-site monitoring points in time step n and the previous L time steps; P(n-1) represents the covariance matrix of the previous time step; λ is the forgetting factor, which is between 0.9 and 1;

[0034] Update the weight vector w(n):

[0035] w(n)=w(n-1)+k(n)[d(n)-x T (n)w(n-1)]

[0036] Wherein, w(n-1) is the iterative update value of W(0); w(n) is the weight vector of the adaptive filter, which represents the parameters of the filter at the nth time step; d(n) is the desired output, that is, the target signal under noise-free conditions;

[0037] Update the inverse of the correlation matrix P(n):

[0038]

[0039] Apply filter w(n) to input x(n) to get filtered output y(n):

[0040] y(n)=w T (n)x(n).

[0041] Furthermore, in step (4), the method for training the neural network mask is: converting the short-time Fourier transform (STFT) of the training set data into a time-frequency spectrum diagram; and sending the time-frequency spectrum diagram into the FRCRN model for cIRM target prediction to train and obtain the mask.

[0042] Furthermore, in step (4), the method of processing the sound signal in the mask frequency domain is:

[0043] The output after filtering in step (3) is converted into a time-frequency spectrum through STFT, the time-frequency spectrum is multiplied with the trained mask, the multiplication result is inversely short-time Fourier transformed, and the enhanced spectrum is converted back to a time domain signal output to eliminate the residual noise signal.

[0044] An echo cancellation and noise reduction system for environmental sound in a steel plant workshop, the echo cancellation and noise reduction system comprising:

[0045] Sound signal sampling module: real-time acquisition of sound signals of on-site monitoring points in the steel plant workshop environment during operation; the on-site monitoring points include various noise sources and various target sources;

[0046] Delay compensation module: aligns the original sound signal and the workshop echo signal from the same on-site monitoring point in time to obtain the aligned signal;

[0047] Linear filter module: The aligned signal output by the delay compensation module is filtered through a linear filter to suppress the noise and echo signals superimposed in the channel;

[0048] Residual noise suppression module based on deep neural network: used to train the neural network mask, use the mask frequency domain to process the sound signal output by the linear filter module to suppress the residual noise signal, and output a clean sound signal.

[0049] Furthermore, the sound signal sampling module adopts a network microphone, and the sound signal sampling module converts the collected sound signal into a digital signal.

[0050] Beneficial technical effects of the present invention:

[0051] The present invention provides an echo denoising method for environmental sound in a steel plant workshop, which is used to eliminate workshop echoes while suppressing noise interference, thereby enhancing the environmental acoustic quality of the steel plant workshop. The method can utilize a microphone to automatically sample and process on-site audio in real time online, and can achieve rapid detection through multi-threaded parallel computing, thereby ensuring the real-time nature of the detection.

[0052] The present invention provides an echo noise reduction system for environmental sounds in a steel plant workshop, which overcomes the audio distortion in existing solutions and allows operators to hear pure target sounds of interest while providing a solution for producing a clean data set, which can be used for subsequent machine learning to achieve further improvement in the level of automation.

[0053] In summary, the method provided by the present invention can effectively reduce the noise level in the working area, while suppressing echoes and maintaining the clarity of important environmental sounds. Through large-scale on-site sound data training, the network can adapt to various steel mill workshop industrial environments with harsh acoustic conditions. The implementation of the system is conducive to the efficient processing of steel mill noise and improves the acoustic quality of the steel mill environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a structural diagram of a time delay compensation module based on generalized cross-correlation in an embodiment of the present invention.

[0055] Figure 2 It is the correlation detection result in the embodiment of the present invention.

[0056] Figure 3 Schematic diagram of the frame division operation in an embodiment of the present invention.

[0057] Figure 4 4 is a structural diagram of a residual noise suppression module based on a deep neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0060] Embodiment 1: A method for echo cancellation and noise reduction of environmental sound in a steel plant workshop, the method comprising the following steps:

[0061] (1) Real-time acquisition of sound signals of on-site monitoring points in a steel mill workshop environment during operation; the on-site monitoring points include various noise sources and various target sources; specifically, by arranging network microphones at various noise sources and various target sources in the steel mill workshop environment, the sound signals of the on-site monitoring points in the operation process are transmitted to a network recording host in real time through the network, so as to achieve storage, playback, and analysis on a computer;

[0062] (2) Aligning the original acoustic signal and the workshop echo signal from the same on-site monitoring point in time to obtain an aligned signal;

[0063] (3) filtering the aligned signal obtained in step (2) to suppress the noise and echo signals superimposed in the channel;

[0064] (4) training the neural network mask, filtering the sound signal processed by the mask frequency domain processing step (3) to suppress the residual noise signal, and outputting a clean sound signal.

[0065] In step (2) of this embodiment, a generalized cross-correlation and phase transformation algorithm is used to calculate the time delay of the original sound signal and the reverberation of the workshop echo signal generated by the on-site workshop.

[0066] In this embodiment, step (2) specifically includes:

[0067] Step (2) specifically includes:

[0068] (2.1) Use the GCC-PHAT algorithm to define generalized cross-correlation, calculate the statistical correlation between the original sound signal of the steel plant operation workshop and the workshop echo signal, and establish the association between the signals; Figure 2The result of correlation detection is shown, and the corresponding position of each highest peak is the compensation point of the corresponding echo. At each time step, the cross-correlation value between the original sound signal of the steel plant operation workshop and the workshop echo signal is updated; wherein, when establishing the association between the signals, in order to achieve real-time calculation, a smoothing parameter is used for the alignment process of the steel plant workshop echo and the target sound signal to balance the signal data at the current moment and the past moment. The cross-correlation value is smoothed by the weighted averaging method to adapt to the dynamic changes of the signal; the GCC-PHAT algorithm used in the present invention is a conventional time delay compensation algorithm, and the present invention only uses the weighted averaging method to make it meet the real-time requirements.

[0069] (2.2) Convert the frequency domain signal back to the time domain signal through inverse fast Fourier transform; in the time domain signal, find the maximum position of the cross-correlation value, calculate the delay time between the signals, and adjust the workshop echo signal according to the delay time so that the workshop echo signal is aligned in time with the corresponding original sound signal.

[0070] Step (2.2) specifically includes: performing fast Fourier transform on the original sound signal x1(t) and the workshop echo signal x2(t) from the same on-site monitoring point to obtain X1(f) and X2(f);

[0071] Cross power spectral density of two signals The calculation formula is:

[0072]

[0073] in, is the conjugate of X2(f);

[0074] Apply phase transformation to obtain the regularized cross power spectral density

[0075]

[0076] The regularized cross power spectral density Perform an inverse fast Fourier transform to convert from the frequency domain to the time domain, obtaining the cross-correlation function

[0077]

[0078] Find the location of the peak in the cross-correlation function τ max , position τ max That is, the estimated delay between the original sound signal x1(t) and the workshop echo signal x2(t);

[0079] According to the calculated delay τ max , adjust the time axis of the signal x2(t) so that x1(t) and x2(t+τmax ) are aligned in time to obtain the aligned signal x(t).

[0080] The original acoustic signals and workshop echo signals of all on-site monitoring points are aligned in time to obtain all aligned signals corresponding to all time steps.

[0081] The aligned signals obtained after processing in this step can be more accurately subjected to subsequent noise suppression processing to ensure that all signals reach an optimal synchronization state before entering the linear filter processing step.

[0082] Step (3) of this embodiment specifically includes:

[0083] Taking the aligned signal x(t) as input, the input signal is first framed and windowed; Figure 3 As shown in the figure, the signal framing and windowing operation aims to divide the continuous audio signal into several 1s frames so as to analyze and process the local signal. The frames are partially overlapped by setting the frame shift to ensure the continuity between frames, reduce the boundary effect caused by framing, and enhance the capture of signal details. Applying a window function to each frame signal, in this case a Hamming window, can reduce spectrum leakage and optimize the frequency domain analysis effect.

[0084] Initialize the filter weight vector w(0) and the inverse of the correlation matrix P(0). Usually P(0) is set to a large scalar multiplied by the unit matrix, such as δI, where δ is a large positive number. Specifically, δ can be set to 10^6 to enhance the adaptability of the algorithm, which is suitable for situations where the input signal has large dynamic changes and requires faster convergence.

[0085] Calculate the gain vector k(n):

[0086]

[0087] Among them, the input vector of each time step n is x(n), x(n) means the alignment signal of all on-site monitoring points in time step n and the previous L time steps, x(n) is the discrete signal after sampling; P(n-1) represents the covariance matrix of the previous time step; λ is the forgetting factor, which is between 0.9 and 1. The forgetting factor is used to adjust the algorithm's memory of old data; x T (n) is the transposed form of the input vector.

[0088] Update the weight vector w(n):

[0089] w(n)=w(n-1)+k(n)[d(n)-x T (n)w(n-1)]

[0090] Wherein, w(n-1) is the iterative update value of W(0); w(n) is the weight vector of the adaptive filter, which represents the parameters of the filter at the nth time step; d(n) is the desired output, that is, the target signal under noise-free conditions;

[0091] Update the inverse of the correlation matrix P(n):

[0092]

[0093] Among them, by recursively updating P(n), the weight vector can be adjusted according to the new input signal, so that the filter can better adapt to the data; P(n-1) is the inverse of the correlation matrix at the previous moment, which contains the statistical information of the input signal at the previous time step; P(n) is an update to P(n-1), which uses the information update that discards the historical data part;

[0094] Apply filter w(n) to input x(n) to get filtered output y(n):

[0095] y(n)=w T (n)x(n).

[0096] In this step, the correlation matrix and gain vector of the input echo reverberation signal are calculated at each time step; the filter coefficients are updated by minimizing the error using the recursive least squares algorithm; and the filter coefficients are iteratively updated at each time step based on the current correlation matrix and correlation vector. The updated filter coefficients are applied to filter the time-aligned near-end signal to extract the target signal after noise suppression. The delayed far-end signal is weighted by the filter coefficients to compensate for the influence of the noise path, and the near-end target signal after filtering is normalized and enhanced. Through the above steps, the wRLS linear filter can effectively suppress the noise in the steel plant environment, provide a clear sound signal, and provide a clean input for subsequent deep learning processing.

[0097] Moreover, in the process of updating the linear filter of the reverberation signal in the workshop, the semi-blind source separation method is used to improve the adaptability of the filter to different noise environments. Different components in the signal are separated through independent component analysis.

[0098] In step (4) of this embodiment, the method for training the neural network mask is: converting the training set data into a time-frequency spectrum diagram by short-time Fourier transform (STFT); and sending the time-frequency spectrum diagram into the FRCRN model for cIRM target prediction to obtain the mask by training.

[0099] In step (4) of this embodiment, the method of processing the sound signal in the mask frequency domain is:

[0100] The output after filtering in step (3) is converted into a time-frequency spectrum through STFT, the time-frequency spectrum is multiplied with the trained mask, the multiplication result is inversely short-time Fourier transformed, and the enhanced spectrum is converted back to a time domain signal output to eliminate the residual noise signal.

[0101] Embodiment 2: An echo cancellation and noise reduction system for environmental sound in a steel plant workshop, the echo cancellation and noise reduction system comprising:

[0102] Sound signal sampling module: used to obtain the sound signals of the on-site monitoring points in the steel plant workshop environment during the working process in real time;

[0103] Delay compensation module: used to align the echo signals from different sound sources in time in a workshop with multiple sound sources; Figure 2 As shown, the delay compensation module uses the generalized cross-correlation and phase transform (GCC-PHAT) algorithm to process multi-source signals, estimate and compensate for the time delay between the sound sources. This module ensures that the signals of different sound sources can be aligned in time, laying the foundation for efficient noise suppression. This module performs a fast Fourier transform on each signal and calculates the cross-power spectral density of the two signals. Apply a phase transform and regularize the cross-power spectral density. Perform an inverse fast Fourier transform to convert from the frequency domain to the time domain to obtain the cross-correlation function. Find the location of the peak in the cross-correlation function τ max This position is the estimated delay between the two signals. max , adjust the time axis of the signal so that the two signals are aligned in time. The aligned signal can be more accurately processed for subsequent noise suppression. This ensures that all signals reach the optimal synchronization state before entering the linear filter processing step.

[0104] Linear filter module: The aligned signal output by the delay compensation module is processed by a linear filter to suppress the superimposed noise and echo signals in the channel; the linear filter module applies weighted recursive least squares filtering technology to adjust the filter coefficients in real time to suppress the superimposed noise and echo signals. This module is particularly suitable for dynamically changing industrial noise environments and can automatically optimize the filter performance according to the real-time changes of environmental noise. The linear filter module also adds a window to the signal and uses the signal aligned by the delay compensation module as input. It includes the weights of the target signal and the noise signal after the initialization filter and the inverse of the correlation matrix, which is usually set to a large scalar multiplied by the unit matrix, such as δI, where δ is a large positive number. For each time step n, consider the input vector x(n), which includes samples of the current and previous L time steps, and calculate the gain vector, update the weight vector, obtain the target signal under noise-free conditions, and update the inverse of the correlation matrix. Apply the filter to the input to obtain the filtered output. Through the above steps, the linear filter module can effectively suppress the noise in the steel plant environment, provide a clear sound signal, and provide a clean input for subsequent deep learning processing.

[0105] Residual noise suppression module based on deep neural network: used to train neural network masks, use mask frequency domain to process sound signals to suppress residual noise signals, and output clean sound signals. Among them, the input signal of the residual noise suppression module based on deep neural network is the sound signal output by the linear filter module. The deep learning training adopts the frequency recursive convolutional recurrent network model (FRCRN), which performs complex ideal ratio mask prediction in the time-frequency domain. By multiplying the mask with the spectrogram of the workshop ambient noise, and converting the enhanced spectrogram back to the time domain signal output through the inverse short-time Fourier transform. This module uses deep learning technology to further suppress residual noise. FRCRN model training uses the time-frequency domain complex ideal ratio mask (cIRM) to effectively extract the target sound signal and convert it back to the time domain signal through the inverse short-time Fourier transform. As Figure 4 As shown in the figure, the module is based on the residual noise suppression method of deep neural network. First, the historical sound signal is converted into a time-frequency graph and decomposed into the real and imaginary parts of the signal. Then, these signals are processed by the FRCRN neural network model to generate an ideal ratio mask. Finally, the time-frequency graph of the real-time sound signal is combined to output the time-frequency graph of the clean sound signal, thereby achieving effective suppression of noise and clear extraction of the target sound.

[0106] In this embodiment, the sound signal sampling module uses a network microphone. By arranging a network microphone near the sound, the sound signal of the on-site shearing process is picked up and transmitted to the network recording host in real time through the network. The sound signal sampling module converts the collected sound signal into a digital signal to achieve storage, playback, and analysis on the computer side. The sound signal sampling module collects sound signals from various noise sources including mechanical operation sound, impact sound, etc. in real time, as well as sound signals from target sources such as fertilizer falling from the shears.

[0107] The echo noise reduction system for a steel plant workshop environment provided by the present invention is intended to effectively reduce the noise level in the working area, while suppressing the echo to maintain the clarity of important environmental sounds. The system integrates four modules: a sound signal sampling module, a time delay compensation module, a linear filtering module, and a residual noise suppression module based on a deep neural network. The sound signal sampling module collects live sound signals in real time through a microphone. The time delay compensation module is used to time align the sounds and echoes generated by multiple sound sources in the steel plant to ensure that all sound signals can be synchronized before entering the subsequent processing module. Through the linear filter module, its filtering parameters are adjusted to adapt to the unique sound characteristics of the steel plant, reduce noise interference, enhance the sounds that need to be paid attention to in the workshop environment, and improve the overall quality of the sound signal. The residual noise suppression module is specially designed to further suppress the residual noise that still exists after filtering by the linear filter module. Through large-scale field sound data training, the network can adapt to various industrial environments of steel plant workshops with harsh acoustic conditions. The system provided by the present invention realizes efficient processing of steel plant noise and improves the acoustic quality of the steel plant environment.

[0108] In a specific embodiment, for broadband speech input, a window length of 20ms and a frame shift of 10ms are used. The total delay of the broadband FRCRN model is limited to less than 30ms; in order to improve the spectral resolution, the STFT length is set to 640, and zero padding technology is used. The number of input channels of the model is 1. The filtered output obtained by the linear filter is converted into a time-frequency diagram through STFT and multiplied with the mask; the result is an inverse short-time Fourier transform to convert the enhanced spectrum back to the time domain signal output to eliminate the residual noise signal.

[0109] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. For ordinary technicians in this field, they can still modify or transform the technical solutions described above, which all fall within the scope of protection of the present invention.

Claims

1. A method for echo cancellation and noise reduction of environmental sound in a steel plant workshop, characterized in that: The method comprises the following steps: (1) Real-time acquisition of sound signals of on-site monitoring points in a steel plant workshop environment during operation; the on-site monitoring points include various noise sources and various target sources; (2) Aligning the original acoustic signal and the workshop echo signal from the same on-site monitoring point in time to obtain an aligned signal; (3) filtering the aligned signal obtained in step (2) to suppress the noise and echo signals superimposed in the channel; (4) training the neural network mask, filtering the sound signal processed by the mask frequency domain processing step (3) to suppress the residual noise signal, and outputting a clean sound signal.

2. According to claim 1, a method for echo cancellation and noise reduction of environmental sound in a steel plant workshop, characterized in that: In step (2), a generalized cross-correlation and phase transformation algorithm is used to calculate the time delay of the original sound signal and the reverberation of the workshop echo signal generated by the on-site workshop.

3. The echo cancellation and noise reduction method for environmental sound in a steel plant workshop according to claim 2 is characterized in that: Step (2) specifically includes: (2.1) Define generalized cross-correlation, calculate the statistical correlation between the original sound signal of the steel plant operation workshop and the echo signal of the workshop, and establish the association between the signals; at each time step, update the cross-correlation value between the original sound signal of the steel plant operation workshop and the echo signal of the workshop; (2.2) The frequency domain signal is converted back to the time domain signal through inverse fast Fourier transform; in the time domain signal, the maximum position of the cross-correlation value is found, the delay time between the signals is calculated, and the workshop echo signal is adjusted so that the workshop echo signal is aligned in time with the corresponding original sound signal.

4. The echo cancellation and noise reduction method for environmental sound in a steel plant workshop according to claim 3 is characterized in that: Step (2.2) specifically includes: Perform fast Fourier transform on the original sound signal x1(t) and the workshop echo signal x2(t) from the same on-site monitoring point to obtain X1(f) and X2(f); Cross power spectral density of two signals The calculation formula is: in, is the conjugate of X2(f); Apply phase transformation to obtain the regularized cross power spectral density The regularized cross power spectral density Perform an inverse fast Fourier transform to convert from the frequency domain to the time domain, obtaining the cross-correlation function Find the location of the peak in the cross-correlation function τ max , position τ max That is, the estimated delay between the original sound signal x1(t) and the workshop echo signal x2(t); According to the calculated delay τ max , adjust the time axis of the signal x2(t) so that x1(t) and x2(t+τ max ) are aligned in time to obtain the aligned signal x(t).

5. The echo cancellation and noise reduction method for environmental sound in a steel plant workshop according to claim 4, characterized in that: In step (3), a weighted recursive least squares filter is used for filtering.

6. The echo cancellation and noise reduction method for environmental sound in a steel plant workshop according to claim 5, characterized in that: Step (3) specifically includes: Take the aligned signal x(t) as input, initialize the filter weight vector w(0) and the inverse of the correlation matrix P(0); Calculate the gain vector k(n): Among them, the input vector of each time step n is x(n), x(n) means the alignment signal of all on-site monitoring points in time step n and the previous L time steps; P(n-1) represents the covariance matrix of the previous time step; λ is the forgetting factor, which is between 0.9 and 1; Update the weight vector w(n): w(n)=w(n-1)+k(n)[d(n)-x T (n)w(n-1)] Wherein, w(n-1) is the iterative update value of W(0); w(n) is the weight vector of the adaptive filter, which represents the parameters of the filter at the nth time step; d(n) is the desired output, that is, the target signal under noise-free conditions; Update the inverse of the correlation matrix P(n): Apply filter w(n) to input x(n) to get filtered output y(n): y(n)=w T (n)x(n)。 7. The echo cancellation and noise reduction method for environmental sound in a steel plant workshop according to claim 1, characterized in that: In step (4), the method for training the neural network mask is: converting the training set data into a time-frequency spectrum by short-time Fourier transform; sending the time-frequency spectrum to the FRCRN model for cIRM target prediction to train and obtain the mask.

8. The echo cancellation and noise reduction method for environmental sound in a steel plant workshop according to claim 3, characterized in that: In step (4), the method of processing the sound signal in the mask frequency domain is: The output after filtering in step (3) is converted into a time-frequency spectrum through STFT, the time-frequency spectrum is multiplied with the trained mask, the multiplication result is inversely short-time Fourier transformed, and the enhanced spectrum is converted back to a time domain signal output to eliminate the residual noise signal.

9. An echo cancellation and noise reduction system for environmental sound in a steel plant workshop, characterized in that: The echo cancellation and noise reduction system comprises: Sound signal sampling module: real-time acquisition of sound signals of on-site monitoring points in the steel plant workshop environment during operation; the on-site monitoring points include various noise sources and various target sources; Delay compensation module: aligns the original sound signal and the workshop echo signal from the same on-site monitoring point in time to obtain the aligned signal; Linear filter module: The aligned signal output by the delay compensation module is filtered through a linear filter to suppress the noise and echo signals superimposed in the channel; Residual noise suppression module based on deep neural network: used to train the neural network mask, use the mask frequency domain to process the sound signal output by the linear filter module to suppress the residual noise signal, and output a clean sound signal.

10. The echo cancellation and noise reduction system for environmental sound in a steel plant workshop according to claim 9, characterized in that: The sound signal sampling module adopts a network microphone, and the sound signal sampling module converts the collected sound signal into a digital signal.

Images