CN117037836B - Real-time sound source separation method and device based on signal covariance matrix reconstruction - Google Patents

Abstract

A real-time sound source separation method and device based on signal covariance matrix reconstruction. The method includes: when multiple sound source signals are detected, the exponential smoothing method is used to calculate the covariance matrix of the mixed signal; and the covariance matrix of the mixed signal is characterized Value decomposition, using eigenvalues to calculate the noise power of different frequency components; using the subspace and theoretical steering vectors composed of the main eigenvalue vectors to calculate the steering vector of the sound source; using the eigenvalue vectors and eigenvalues of the mixed signal covariance matrix, calculating The inverse matrix of the mixed signal covariance matrix; use the inverse matrix of the mixed signal covariance matrix and the steering vector of the sound source to calculate the power of each sound source, and reconstruct the signal covariance of each sound source according to the definition of the covariance matrix Matrix; use the signal covariance matrix of each sound source and the theoretical guidance vector of the signal to calculate the separation coefficient matrix; obtain the separated sound source signal based on the mixed sound signal vector and the separation coefficient matrix.

Description

Real-time sound source separation method and device based on signal covariance matrix reconstruction

Technical field

The invention relates to the field of array sound source signal processing, and in particular to a real-time sound source separation method and device based on signal covariance matrix reconstruction.

Background technique

The human ear’s ability to extract the target speaker’s voice in complex acoustic scenes with background noise, reverberation, and interference from multiple speakers is known as the “cocktail party effect.” However, this is a difficult task for robots and a difficult problem in the field of sound source signal processing. The technology of extracting the sound source signals of one or more target speakers from the mixed sound source signals using signal processing methods is called sound source separation technology.

Sound source separation mainly includes single-channel sound source separation and multi-channel sound source separation. The multi-channel sound source separation method based on microphone array can better utilize the spatial information of the sound field. Therefore, whether using traditional signal processing methods or deep learning methods, multi-channel sound source separation methods can usually achieve better performance than single-channel sound source separation. At present, most human-computer interaction devices use multi-channel microphone arrays as sound reception hardware.

Theoretically, beamforming based on multi-channel microphone arrays can adaptively complete the separation and extraction of sound source signals in different directions. However, for scenes where multiple people are talking at the same time, it is still difficult to accurately estimate the covariance matrix of interference signals and noise signals composed of other sound source signals for target sound source signals in different directions. In addition, accurate guidance vectors cannot be obtained due to errors such as scattering and reflection in the sound field, microphone position, and sound source localization. These two problems directly affect the sound source separation performance of the beamforming method.

In order to estimate the signal covariance matrix more accurately, one method is to introduce signal presence probability (SignalPresence Probability, SPP) estimation. When the estimated SPP of the target sound source is low, the signal covariance matrix is updated; when the estimated SPP of the target sound source is high, the signal covariance matrix is not updated. However, this method needs to know the a priori probability of the existence of sound source signals, and for scenarios where multiple sound sources exist at the same time, it is difficult to accurately estimate the SPP of different sound source signals.

Another method to estimate the signal covariance matrix is to integrate the Capon spectrum estimate in the entire space area except the angular area where the target sound source signal is located, perform eigenvalue decomposition on the covariance matrix obtained by integration, and reconstruct the signal covariance again. matrix. Although the integral can be discretized into a summation to reduce the amount of calculation, for wide-band sound source signals, multiple summation operations, inversion operations, and eigenvalue decomposition operations need to be performed at each frequency, and the computational complexity is very high. It cannot meet the requirements of real-time sound source signal processing.

Contents of the invention

In view of the shortcomings of the existing technology, the present invention proposes a real-time sound source signal separation method and device based on signal covariance matrix reconstruction. Estimate the noise power through eigenvalue decomposition, use eigenvalue decomposition to find the inverse matrix, use the projection of the theoretical guidance vector in the subspace composed of the main eigenvalue vectors to correct the guidance vector of the sound source, and reconstruct each sound according to the definition of the covariance matrix. The covariance matrix of the source not only greatly reduces the calculation amount of signal covariance matrix reconstruction, but also significantly improves the reconstruction accuracy of the covariance matrix.

The first aspect of the present invention provides a sound source separation method based on signal covariance matrix reconstruction, including the following steps:

When multiple sound source signals are detected, the exponential smoothing method is used to calculate the covariance matrix of the mixed signal;

Perform eigenvalue decomposition on the covariance matrix of the mixed signal, and use the eigenvalues to calculate the noise power of different frequency components;

Calculate the guidance vector of the sound source using the subspace composed of the main eigenvalue vectors and the theoretical guidance vector;

Calculate the inverse matrix of the mixed signal covariance matrix using the eigenvalue vector and eigenvalue of the mixed signal covariance matrix;

Use the inverse matrix of the mixed signal covariance matrix and the steering vector of the sound source to calculate the power of each sound source, and reconstruct the signal covariance matrix of each sound source according to the definition of the covariance matrix;

The signal covariance matrix of each sound source and the theoretical steering vector of the signal are used to calculate the separation coefficient matrix; the separated sound source signal is obtained based on the mixed sound signal vector and the separation coefficient matrix.

Furthermore, when multiple sound source signals are detected, the formula for calculating the covariance matrix of the mixed signal using the exponential smoothing method is as follows:

(1)

in Represents the time frame;/> Represents frequency;/> It is an exponential smoothing factor, with a value range of (0, 1). The larger the value, the greater the role of historical forecast data, and the smaller the role of current actual data;

(2)

for Dimensional sound signal vector;/> For the first/> Signals observed by microphones;/> is the transpose of a regular matrix; and when/> When,/> for:

(3)

in for/> Order identity matrix;/> It is a regularization parameter that changes with frequency. The value range is (0, 1). The greater the frequency, the smaller the value;/> is the covariance matrix of spherical diffusion noise, whose/> The elements are:

(4)

in is the delay of adjacent microphone array elements,/> is the spacing between adjacent microphone array elements,/> is the propagation speed of physical quantity,/> is the sampling function.

Furthermore, the covariance matrix of the mixed signal is decomposed into eigenvalues, and the process of using eigenvalues to calculate the noise power of different frequency components includes:

Mixed-signal covariance matrix for the current time frame Do eigenvalue decomposition:

(5)

in is the eigenvalue vector matrix;/> It is a diagonal matrix, and its diagonal elements are the eigenvalues arranged from large to small; is the conjugate transpose of the matrix.

Calculate the power of noise is the smallest/> The average of eigenvalues, where/> is the total number of sound sources.

Furthermore, the guidance vector of each sound source is calculated using the subspace composed of the main eigenvalue vectors and the theoretical guidance vector. The formula is as follows:

(6)

in For the first/> The direction of the incoming wave of a sound source, in the three-dimensional spherical coordinate system,/> ,/> and/> are the pitch angle and azimuth angle respectively;/> For the first/> The theoretical guidance vector of each sound source is calculated based on the topology of the array and the free sound field propagation model;/> for/> The front/> composed of column vectors/> dimensional matrix.

Furthermore, using the eigenvalue vector and eigenvalues, the formula for calculating the inverse matrix of the mixed signal covariance matrix is as follows:

(7)

in for/> Obtained by taking the reciprocal of the diagonal elements of .

Furthermore, the process of calculating the power of each sound source using the inverse matrix of the mixed signal covariance matrix and the steering vector of the sound source, and reconstructing the signal covariance matrix of each sound source according to the definition of the covariance matrix includes:

Calculate the first The power of a sound source/> The formula is:

(8)

According to the definition of covariance matrix, reconstruct the Signal covariance matrix of sound sources/> The formula is as follows:

(9)

Furthermore, the signal covariance matrix of each sound source and the theoretical guidance vector of the target signal are used to calculate the separation coefficient matrix; the process of obtaining the separated sound source signal based on the mixed sound signal vector and the separation coefficient matrix includes:

calculate The formula of the dimensional separation coefficient matrix is as follows:

(10)

in for/> composed of theoretical guidance vectors of sound sources/> dimensional matrix, /> is the signal covariance matrix ultimately used to calculate the separation coefficient:

(11)

in Load the weight factor for the diagonal, the value range is (0,1), the larger the value, the more sensitive it is to signal changes; where:

(12)

In the above formula, calculate the When the separation coefficient of a sound source is Excludes No./> Signal covariance matrix of sound sources/> .

The formula for calculating the sound source separation of the mixed signal received by the microphone is as follows:

(13)

Further, the sound source separation method also includes: separating the frequency domain signal of each frame Multiply by the custom window function and perform inverse Fourier transform to get/> separated time domain acoustic signal.

A second aspect of the present invention provides a sound source separation device based on signal covariance matrix reconstruction, including one or more processors for implementing the above-mentioned real-time sound source separation method of signal covariance matrix reconstruction.

A third aspect of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, it is used to implement the above-mentioned sound source separation method based on signal covariance matrix reconstruction.

The beneficial effects of the present invention are: the present invention proposes a real-time sound source separation method for signal covariance matrix reconstruction. The noise power is estimated through eigenvalue decomposition, and the inverse matrix is obtained using eigenvalue decomposition. The projection of the theoretical steering vector in the subspace composed of the main eigenvalue vectors is used to correct the steering vector of the sound source, which not only greatly reduces the signal covariance matrix repetition The calculation amount of the structure and the reconstruction accuracy of the covariance matrix are suitable for the real-time extraction of different sound source signals by the robot.

Description of the drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without exerting any creative effort.

Figure 1 is a flow chart of the real-time sound source separation method for signal covariance matrix reconstruction proposed by the present invention;

Figure 2 is a comparison chart between the speech separation performance indicators of the method of the present invention and the speech separation performance indicators of other methods;

Figure 3(a) and Figure 3(b) are waveform comparison diagrams of the reference target speech signal and the target speech signal separated from the mixed signal by the method of the present invention, where Figure 3(a) is the waveform diagram of the reference target speech signal, Fig. 3(b) is the waveform diagram of the target speech signal separated from the mixed signal;

Figure 4(a) and Figure 4(b) are waveform comparison diagrams of the reference interference signal and the interference signal suppressed by speech separation according to the method of the present invention. Figure 4(a) is the waveform diagram of the reference interference signal, and Figure 4(b) is the waveform diagram of the reference interference signal. The waveform of the interference signal after suppression by the speech separation method of the present invention;

Figure 5 is a schematic structural diagram of the real-time sound source separation method device for signal covariance matrix reconstruction proposed by the present invention.

Detailed ways

The implementation steps and beneficial effects of the present invention will be described in detail below with reference to the accompanying drawings and specific examples.

Example 1

The present invention proposes a real-time sound source separation method for signal covariance matrix reconstruction, as shown in Figures 1 and 2. The method specifically includes the following steps:

1. Calculate the covariance matrix of the mixed signal ;

The number of known sound sources is , No./> The incoming wave direction vector of a sound source is/> , of which/> and/> are the pitch angle and azimuth angle of the three-dimensional spherical coordinate system respectively, and the number of elements of the microphone array is/> .

The microphone array element receives the Frame time domain signal, for/> Fast Fourier transform of points. For each frequency component, calculate according to the following formula/> Covariance matrix of dimensional mixed signals/> :

(1)

in Represents the time frame;/> Represents frequency;/> It is an exponential smoothing factor with a value range of (0,1). The larger the value, the greater the role of historical forecast data and the smaller the role of current actual data;

(2)

for Dimensional sound signal vector;/> For the first/> signal from a microphone;/> is the transpose of a regular matrix; when When,/> for:

(3)

in for/> Order identity matrix;/> It is a regularization parameter that changes with frequency. The value range is (0,1). The greater the frequency, the smaller the value;/> is the covariance matrix of spherical diffusion noise, whose/> The elements are:

(4)

in is the delay of adjacent microphone array elements,/> is the spacing between adjacent microphone array elements,/> is the propagation speed of physical quantities.

2. Calculate the power of noise ;

Mixed-signal covariance matrix for the current time frame Do eigenvalue decomposition:

(5)

in for/> dimensional eigenvalue vector matrix;/> is a diagonal matrix, and its diagonal elements are the eigenvalues arranged from large to small;/> is the conjugate transpose of the matrix.

Calculate the power of noise is the smallest/> the average of the eigenvalues.

3. Calculate the guidance vector of each sound source using the subspace composed of the main eigenvalue vectors and the theoretical guidance vector. :

(6)

in For the first/> The direction of the incoming wave of a sound source, in the three-dimensional spherical coordinate system,/> ,/> and/> are the pitch angle and azimuth angle respectively;/> For the first/> The theoretical guidance vector of each sound source is calculated based on the topology of the array and the free sound field propagation model;/> for/> The front/> composed of column vectors/> dimensional matrix.

4. Use the following formula to calculate the inverse matrix of the mixed signal covariance matrix:

(7)

in for/> Obtained by taking the reciprocal of the diagonal elements of .

5. Use the following formula to calculate the The power of a sound source/> :

(8)

According to the definition of covariance matrix, reconstruct the Signal covariance matrix of sound sources/> : (9).

6. Using the signal covariance matrix of each sound source and the theoretical guidance vector of the target signal, calculate Dimensional separation coefficient matrix:

(10)

in for/> composed of theoretical guidance vectors of sound sources/> dimensional matrix, /> is the signal covariance matrix ultimately used to calculate the separation coefficient:

(11)

in Load the weight factor for the diagonal, the value range is (0,1), the larger the value, the more sensitive it is to signal changes; where:

(12)

In the above formula, calculate the When the separation coefficient of a sound source is Excludes No./> Signal covariance matrix of sound sources/> .

The formula for calculating the sound source separation of the mixed signal received by the microphone is as follows: (13).

7. Separate the frequency domain signal of each frame Multiply by the custom window function and perform inverse Fourier transform to get/> separated time domain acoustic signal.

Taking a uniform annular six-element microphone array as an example, the microphone is omnidirectional. , the pitch angles of the microphone array elements are all 90°, that is, the torus is installed horizontally.

The microphone array element receives the Frame time domain signal, for/> Fast Fourier transform of points, in this embodiment,/> . According to formula (1), calculate the /> of each frequency component Covariance matrix of dimensional mixed signals/> . In this embodiment,/> . Calculate the initial signal covariance matrix according to formulas (3) and (4) .

In this embodiment, the sampling frequency is 1.6 kHz, and the frequency The range is 0 to 8000 Hz, with values at equal intervals of 62.5 Hz;/> The value of with frequency/> increases and decreases from 0.01 to 0.001 at equal intervals.

According to formula (5), the mixed signal covariance matrix of the current time frame Perform eigenvalue decomposition and use the following formula to calculate the inverse matrix of the mixed signal covariance matrix/> .

Calculate the power of noise is the smallest/> The average value of the characteristic values, in this embodiment,/> .

Assuming a free sound field far-field propagation model, the theoretical guidance vector of a circular microphone array is:

(14)

in is the wave number,/> is the radius of the microphone array, is the azimuth angle of the microphone array element, and:

(15)

According to formula (6), calculate the Guidance vectors of sound sources/> , and calculate the /> The power of a sound source . Then calculate the /> Signal covariance matrix of sound sources/> .

Calculate according to formula (10)-(12) Dimension separation coefficient matrix, in this embodiment, /> .

Finally, perform sound source separation on the mixed signal received by the microphone according to formula (13), and obtain the separated frequency domain signal of each frame. Multiply by the custom window function and perform inverse Fourier transform to get/> separated time domain acoustic signal. In this example, the custom window function is the Hanning window.

In the embodiment of the present invention, the sound source signals used are two speech signals randomly selected from the corpus. The two voice signals come in different directions, with the pitch angle fixed at 88° and the azimuth angle random, but the interval is maintained at 45°. The length of the two voice signals is fixed to 3 seconds, and the energy ratio of the signals is [0,3] dB randomly. The ImageSource Method is used to simulate the mixed speech signal received by the ring 6 microphone, and the reverberation time of the simulated sound field environment is 0.6 s. A total of 100 mixed speech signals are generated.

Figure 2 is a comparison of speech separation performance indicators achieved by the method of the present invention and other methods, including perceptual speech quality evaluation index (PESQ), short-term objective intelligibility (STOI) and scale-invariant signal-to-distortion ratio improvement (SI-SDRi). ). The value range of PESQ is -0.5~4.5. The higher the PESQ value, the better the auditory speech quality of the tested speech. The value range of STOI is 0~1. The closer to 1, the more fully understood the speech is. SI -SNR represents the closeness of the speech separated signal to the original signal, the bigger the better. The average PESQ, average STOI and average SI-SNR of the method of the present invention are all comparable to the beamforming sound source separation method based on Chinese patent CN105182298A.

Table 1 is a comparison of the average total time consuming for processing 3s data of the method of the present invention and other methods. The total time consuming for processing 3s data can reflect whether a method can be used for real-time processing. When the total processing time is much greater than the data duration of 3 seconds, it cannot be used for real-time sound source separation processing. As can be seen from Table 1, the average total processing time of the method of the present invention is reduced by 93.6% compared with the beamforming sound source separation method based on Chinese patent CN105182298A, and can be used for real-time processing. In this embodiment, the CPU used for calculation is Intel Xeon processor icelake, with a main frequency of 2.6GHz. The number of fast Fourier transform points per frame is 256, the frame shift is 128, the signal sampling frequency is 16 kHz, the covariance matrix is reconstructed every 4 frames of signal and the separation matrix coefficients are updated.

Figure 3(a) and Figure 3(b) are waveform comparison diagrams of the reference target speech signal and the target speech signal separated from the mixed signal by the method of the present invention, where Figure 3(a) is the waveform diagram of the reference target speech signal, Fig. 3(b) is the waveform diagram of the target speech signal separated from the mixed signal; it can be seen that the waveform of the target speech signal separated from the mixed signal is very close to the waveform of the reference target speech signal, and a good speech separation effect is achieved.

Figure 4(a) and Figure 4(b) are waveform comparison diagrams of the reference interference signal and the interference signal suppressed by speech separation according to the method of the present invention. Figure 4(a) is the waveform diagram of the reference interference signal, and Figure 4(b) is the waveform diagram of the reference interference signal. The waveform of the interference signal after being suppressed by the voice separation method of the present invention; it can be seen that the interference voice signal is almost completely suppressed after being processed by the present invention.

Corresponding to the foregoing embodiments of the real-time sound source separation method based on signal covariance matrix reconstruction, the present invention also provides embodiments of a real-time sound source separation method and device based on signal covariance matrix reconstruction.

Example 2

Referring to Figure 5, this embodiment provides a real-time sound source separation method device for signal covariance matrix reconstruction, including one or more processors, used to implement the real-time sound source separation method for signal covariance matrix reconstruction in Embodiment 1. Source separation methods.

The embodiments of the real-time sound source separation method and device for signal covariance matrix reconstruction of the present invention can be applied to any device with data processing capabilities, and any device with data processing capabilities can be a device or device such as a computer. The device embodiments may be implemented by software, or may be implemented by hardware or a combination of software and hardware. Taking software implementation as an example, as a logical device, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory and running them through the processor of any device with data processing capabilities. From the hardware level, as shown in Figure 5, it is a hardware structure diagram of any device with data processing capabilities where the real-time sound source separation method for signal covariance matrix reconstruction of the present invention is located. In addition to the processing shown in Figure 5 In addition to the processor, memory, network interface, and non-volatile memory, any device with data processing capabilities where the device in the embodiment is located may also include other hardware based on the actual functions of any device with data processing capabilities. This will not be described again.

For details on the implementation process of the functions and effects of each unit in the above device, please refer to the implementation process of the corresponding steps in the above method, and will not be described again here.

As for the device embodiment, since it basically corresponds to the method embodiment, please refer to the partial description of the method embodiment for relevant details. The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in One location, or it can be distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Persons of ordinary skill in the art can understand and implement the method without any creative effort.

Example 3

An embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the real-time sound source separation method for signal covariance matrix reconstruction in the above-mentioned Embodiment 1 is implemented.

The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities as described in any of the foregoing embodiments, such as a hard disk or a memory. The computer-readable storage medium can also be any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, and a flash card (Flash Card) equipped on the device. wait. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or is to be output.

Claims (10)

1. A real-time sound source separation method based on signal covariance matrix reconstruction, which is characterized by including the following steps: When multiple sound source signals are detected, the exponential smoothing method is used to calculate the covariance matrix of the mixed signal; Perform eigenvalue decomposition on the covariance matrix of the mixed signal, and use the eigenvalues to calculate the noise power of different frequency components; Calculate the guidance vector of the sound source using the subspace composed of the main eigenvalue vectors and the theoretical guidance vector; Calculate the inverse matrix of the mixed signal covariance matrix using the eigenvalue vector and eigenvalue of the mixed signal covariance matrix; Use the inverse matrix of the mixed signal covariance matrix and the steering vector of the sound source to calculate the power of each sound source, and reconstruct the signal covariance matrix of each sound source according to the definition of the covariance matrix; The signal covariance matrix of each sound source and the theoretical steering vector of the signal are used to calculate the separation coefficient matrix; the separated sound source signal is obtained based on the mixed sound signal vector and the separation coefficient matrix. 2. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 1, characterized in that when multiple sound source signals are detected, the exponential smoothing method is used to calculate the formula of the covariance matrix of the mixed signal. as follows: (1) in Represents the time frame;/> Represents frequency;/> is the exponential smoothing factor, the value range is/> , the larger the value, the greater the role of historical forecast data, and the smaller the role of current actual data; (2) for Dimensional sound signal vector;/> For the first/> Signals observed by microphones;/> is the transpose of a regular matrix; and when/> =1,/> for (3) in for/> Order identity matrix;/> is a regularization parameter that changes with frequency, and its value range is/> , the greater the frequency, the smaller the value;/> is the covariance matrix of spherical diffusion noise, whose/> The elements are: (4) in is the delay of adjacent microphone array elements,/> is the spacing between adjacent microphone array elements,/> is the propagation speed of physical quantity,/> is the sampling function. 3. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 1, characterized in that the covariance matrix of the mixed signal is decomposed into eigenvalues, and the eigenvalues are used to calculate the noise power of different frequency components. The process includes: Mixed-signal covariance matrix for the current time frame Do eigenvalue decomposition: (5) in is the eigenvalue vector matrix;/> is a diagonal matrix, and its diagonal elements are the eigenvalues arranged from large to small;/> is the conjugate transpose of the matrix; Calculate the power of noise ：/> is the smallest/> The average of eigenvalues, where/> is the total number of sound sources. 4. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 1, characterized in that the steering vector of each sound source is calculated using a subspace composed of main eigenvalue vectors and a theoretical steering vector. The formula is as follows: (6) in For the first/> The direction of the incoming wave of a sound source, in the three-dimensional spherical coordinate system,/> ,/> and/> are the pitch angle and azimuth angle respectively;/> For the first/> The theoretical guidance vector of each sound source is calculated based on the topology of the array and the free sound field propagation model;/> for/> before/> composed of column vectors/> dimensional matrix. 5. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 1, characterized in that, using eigenvalue vectors and eigenvalues, the formula for calculating the inverse matrix of the mixed signal covariance matrix is as follows: (7) in for/> Obtained by taking the reciprocal of the diagonal elements of . 6. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 1, characterized in that the power of each sound source is calculated using the inverse matrix of the mixed signal covariance matrix and the steering vector of the sound source. , and the process of reconstructing the signal covariance matrix of each sound source according to the definition of the covariance matrix includes: Calculate the first The power of a sound source/> The formula is: (8) According to the definition of covariance matrix, reconstruct the Signal covariance matrix of sound sources/> The formula is as follows: (9). 7. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 1, characterized in that the separation coefficient matrix is calculated using the signal covariance matrix of each sound source and the theoretical guidance vector of the target signal; The process of obtaining the separated sound source signal based on the mixed sound signal vector and the separation coefficient matrix includes: calculate The formula of the dimensional separation coefficient matrix is as follows: (10) in for/> composed of theoretical guidance vectors of sound sources/> dimensional matrix, /> is the signal covariance matrix ultimately used to calculate the separation coefficient: (11) in Load the weight factor for the diagonal, the value range is/> , the larger the value, the more sensitive it is to signal changes; where (12) In the above formula, calculate the When the separation coefficient of a sound source is Excludes No./> Signal covariance matrix of sound sources/> ; The formula for calculating the sound source separation of the mixed signal received by the microphone is as follows: (13). 8. The real-time sound source separation method based on signal covariance matrix reconstruction according to claim 7, characterized in that the sound source separation method further includes: separating the frequency domain signal of each frame Multiply by the custom window function and perform inverse Fourier transform to get/> separated time domain acoustic signal. 9. A real-time sound source separation device based on signal covariance matrix reconstruction, characterized in that it includes one or more processors for implementing the signal covariance matrix-based method described in any one of claims 1-8. A reconstructed real-time sound source separation method. 10. A computer-readable storage medium, characterized in that a program is stored thereon, and when the program is executed by a processor, it is used to implement the signal covariance matrix reconstruction according to any one of claims 1-8. A real-time sound source separation method.