CN114373475A - Method, device and storage medium for speech noise reduction based on microphone array - Google Patents

Abstract

The present application discloses a voice noise reduction method based on a microphone array, which solves the problem that the complexity of solving the filter in the prior art will increase rapidly with the increase of the filter length, and the statistical characteristics of the voice signal and noise will change. The method includes: acquiring the noisy speech signal; preprocessing the noisy speech signal to determine the frequency domain noisy speech signal; estimating the frequency domain noisy speech signal and the statistical characteristics of the noise signal; The array is divided into multiple sub-arrays, and multiple sub-filters are estimated respectively, and the frequency-domain noise reduction filter is determined; according to the frequency-domain noise reduction filter, the frequency-domain noisy speech signal is denoised and converted into time-domain noise reduction speech signal, so that the required signal covariance matrix dimension is smaller in the process of solving the filter, thereby significantly reducing the complexity of solving the speech noise reduction filter, and improving the filter's sensitivity to changes in the statistical characteristics of speech signals and noise. tracking ability.

Description

Method, device and storage medium for speech noise reduction based on microphone array

technical field

The present application relates to the technical field of microphone arrays, and in particular, to a method, device and storage medium for speech noise reduction based on a microphone array.

Background technique

Voice noise reduction plays a pivotal role in intelligent voice, human-computer interaction, teleconferencing, hearing aids, vehicle, virtual reality, immersive communication, and military voice communication with ultra-high background noise, and its performance is good or bad. It directly affects the experience of voice interaction.

Early voice interaction systems are usually equipped with only one microphone, and the corresponding noise reduction method is single-channel voice noise reduction. The single-channel speech noise reduction method has the advantages of simple implementation and high computing efficiency, and can achieve certain effects, but it also has great limitations. Studies have shown that under certain conditions, single-channel noise reduction will definitely introduce speech distortion, and the greater the improvement of the signal-to-noise ratio, the greater the introduced speech distortion. In contrast, multi-channel speech noise reduction methods have the potential to significantly improve the signal-to-noise ratio with little or no speech distortion. Classical multi-channel speech noise reduction methods include multi-channel Wiener filtering, multi-channel compromise filtering, minimum variance undistorted response filtering, linear constrained minimum variance filtering, and generalized sidelobe cancellation. In recent years, researchers at home and abroad have proposed speech noise reduction methods based on deep learning, which can achieve good performance.

In order to achieve better speech noise reduction performance, it is usually necessary to equip more microphones to obtain richer space-time-frequency information. But this also usually means designing longer filters. The application of longer filters brings the following two problems. First, the complexity of solving the filter will increase rapidly with the increase of the filter length; second, the dimension of the signal covariance matrix required in the process of solving the filter will be larger, so it needs to be more Many observation samples are used to estimate the covariance matrix of the signal, which is used to calculate the coefficient of the filter, which leads to a decrease in the ability to track the changes in the statistical characteristics of the speech signal and noise, and cannot better handle the common non-stationary noise in practice.

SUMMARY OF THE INVENTION

The embodiment of the present application solves two problems caused when the filter length is long in the prior art by providing a voice noise reduction method based on a microphone array, that is, first, the complexity of solving the filter will increase It increases rapidly with the increase of the filter length; secondly, the dimension of the signal covariance matrix required in the process of solving the filter will be larger, so more observation samples are needed to estimate the signal covariance The matrix is used to calculate the coefficients of the filter, which leads to a decrease in the ability to track the changes in the statistical characteristics of the speech signal and noise, and cannot better handle the non-stationary noise that is common in practice. The embodiment of the present application significantly reduces the complexity of solving the filter, and the required signal covariance matrix dimension in the process of solving the filter is smaller, so the covariance matrix can be estimated with fewer signal observation samples, Thereby, the ability of the filter to track the changes of the statistical characteristics of the speech signal and noise is improved.

In a first aspect, an embodiment of the present invention provides a method for noise reduction based on a microphone array, the method comprising:

Obtain noisy speech signal;

Preprocessing the noisy speech signal to determine the frequency domain noisy speech signal;

Estimating the statistical properties of the frequency-domain noisy speech signal, and estimating the statistical properties of the noise signal;

Divide the microphone array into multiple sub-arrays, and estimate multiple sub-filters respectively;

determining a frequency-domain noise reduction filter according to the plurality of sub-filters;

According to the frequency-domain noise reduction filter, noise reduction processing is performed on the frequency-domain noisy speech signal to determine the frequency-domain noise reduction speech signal;

Converting the frequency-domain noise-reduced speech signal into a time-domain noise-reduced speech signal.

With reference to the first aspect, in a possible implementation manner, the preprocessing of the noisy speech signal includes: framing and windowing the noisy speech signal and then performing fast Fourier transform.

With reference to the first aspect, in a possible implementation, the estimating the statistical characteristics of the frequency-domain noisy speech signal includes estimating the statistical characteristics of the noisy speech signal according to a time smoothing estimation method.

With reference to the first aspect, in a possible implementation manner, the estimating the statistical characteristics of the noise signal includes estimating the statistical characteristics of the noise signal according to an existing noise estimation algorithm.

With reference to the first aspect, in a possible implementation manner, dividing the microphone array into multiple sub-arrays and estimating multiple sub-filters respectively includes iteratively estimating multiple sub-filters by using the low-rank structure of the noise reduction filter .

In a second aspect, an embodiment of the present invention provides a voice noise reduction device based on a microphone array, which is characterized by comprising:

The signal acquisition module is used to acquire the noisy speech signal;

a signal preprocessing module for preprocessing the noisy speech signal to determine the frequency domain noisy speech signal;

a statistical characteristic estimation module for estimating the statistical characteristic of the frequency-domain noisy speech signal, and estimating the statistical characteristic of the noise signal;

a sub-filter determining module, used for dividing the microphone array into multiple sub-arrays, and estimating multiple sub-filters respectively;

a frequency-domain noise reduction filter determination module, configured to determine a frequency-domain noise reduction filter according to the plurality of sub-filters;

A noise reduction module, configured to perform noise reduction processing on the frequency-domain noisy speech signal according to the frequency-domain noise reduction filter, to determine the frequency-domain noise reduction speech signal;

A time-domain noise-reduced speech signal determination module, configured to convert the frequency-domain noise-reduced speech signal into a time-domain noise-reduced speech signal.

With reference to the second aspect, in a possible implementation manner, the signal preprocessing module includes: performing fast Fourier transform on the noisy speech signal after framing and windowing.

With reference to the second aspect, in a possible implementation manner, the statistical characteristic estimation module includes: including estimating the statistical characteristic of a noisy speech signal according to a time smoothing estimation method.

With reference to the second aspect, in a possible implementation manner, the statistical characteristic estimation module includes: including estimating the statistical characteristic of the noise signal according to an existing noise estimation algorithm.

With reference to the second aspect, in a possible implementation manner, the frequency-domain noise reduction filter determination module includes: iteratively estimating a plurality of sub-filters by using a low-rank structure of the noise reduction filter.

In a third aspect, an embodiment of the present invention provides a voice noise reduction server based on a microphone array, including a memory and a processor;

the memory for storing computer-executable instructions;

The processor is adapted to execute the computer-executable instructions to implement the method of the first aspect.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores executable instructions, and when a computer executes the executable instructions, any one of the first aspect can be implemented Methods.

One or more technical solutions provided in the embodiment of the present invention have at least the following technical effects or advantages:

The embodiment of the present invention adopts a voice noise reduction method based on a microphone array. The method includes: acquiring a noisy voice signal; preprocessing the noisy voice signal to determine a frequency-domain noisy voice signal; estimating a frequency-domain noisy voice signal Statistical characteristics of the signal, estimate the statistical characteristics of the noise signal; divide the microphone array into multiple sub-arrays, and estimate multiple sub-filters respectively; determine the frequency-domain noise reduction filter according to the multiple sub-filters; According to the frequency-domain noise reduction filter The noise reduction processing is performed on the frequency-domain noise-reduced speech signal to determine the frequency-domain noise-reduced speech signal; the frequency-domain noise-reduced speech signal is converted into a time-domain noise reduction speech signal. It effectively solves two problems caused when the filter length is long in the prior art, namely, first, the complexity of solving the filter will increase rapidly with the increase of the filter length; second, The dimension of the signal covariance matrix required in the process of solving the filter will be larger, so more observation samples are needed to estimate the covariance matrix of the signal to calculate the coefficients of the filter, which will lead to the loss of speech signal and noise. The ability to track changes in statistical properties is reduced, and it cannot better handle non-stationary noise that is common in practice. The embodiment of the present invention significantly reduces the complexity of solving the filter, and the required signal covariance matrix dimension in the process of solving the filter is smaller, so the covariance matrix can be estimated by using fewer signal observation samples, Thereby, the ability of the filter to track the changes of the statistical characteristics of the speech signal and noise is improved.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings that are required in the description of the embodiments of the present invention or the prior art. Obviously, the drawings in the following description These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

Fig. 1 is a flow chart of steps of a microphone array-based voice noise reduction method provided by an embodiment of the application;

2 is a schematic diagram of a device for noise reduction based on a microphone array provided by an embodiment of the present application;

3 is a schematic diagram of a server for voice noise reduction based on a microphone array provided by an embodiment of the present application;

4 is a comparison diagram of the complexity of the method provided by the embodiment of the present application and the complexity of the traditional method;

FIG. 5 is an image of the mean square error of the method provided by the embodiment of the present application as a function of the number of iterations;

FIG. 6 is a comparison diagram of the mean square error of the method provided by the embodiment of the present application and the traditional method when the statistical characteristics of noise suddenly change, provided by the embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work, all belong to the protection scope of the present invention.

In the early voice interaction system, only one microphone is usually equipped, and the corresponding voice noise reduction method is single-channel voice noise reduction. The single-channel speech noise reduction method has the advantages of simple implementation and high computing efficiency, and can achieve certain effects, but it also has great limitations. Studies have shown that under certain conditions, single-channel noise reduction will definitely introduce speech distortion, and the greater the improvement of the signal-to-noise ratio, the greater the introduced speech distortion. In contrast, the multi-channel speech noise reduction method has more potential, and can significantly improve the signal-to-noise ratio under the premise of introducing little or no speech distortion. Multi-channel speech noise reduction usually requires more microphones to obtain richer space-time-frequency information. However, it will lead to two problems. First, the complexity of solving the filter will increase rapidly with the increase of the filter length; second, the signal covariance matrix required in the process of solving the filter will increase. The dimension is larger, so more test samples are needed to estimate the covariance matrix of the signal to calculate the coefficients of the filter, which leads to a decrease in the tracking ability of the statistical changes of the speech signal and noise, which cannot be better processed in practice. Common non-stationary noise.

An embodiment of the present invention provides a voice noise reduction method based on a microphone array. As shown in FIG. 1 , the method includes the following steps:

Step S101, acquiring a noisy speech signal.

Step S102, preprocessing the noisy speech signal to determine the frequency domain noisy speech signal.

Step S103, estimating the statistical characteristics of the frequency-domain noisy speech signal, and estimating the statistical characteristics of the noise signal.

Step S104: Divide the microphone array into a plurality of sub-arrays, and estimate a plurality of sub-filters respectively.

Step S105: Determine a frequency-domain noise reduction filter according to a plurality of sub-filters.

Step S106, performing noise reduction processing on the frequency-domain noisy speech signal according to the frequency-domain noise reduction filter to determine the frequency-domain noise reduction speech signal.

Step S107, converting the frequency-domain noise reduction speech signal into a time-domain noise reduction speech signal.

Combining the above method steps, a more reasonable filter is constructed, which avoids calculating a long filter as a whole like the existing multi-channel speech noise reduction method, and a shorter filter means less filter coefficients. Therefore, compared with the existing method, the method provided by the present application significantly reduces the complexity of solving the speech noise reduction filter, and the required signal covariance matrix dimension in the process of solving the filter is small, so it can be used Fewer signal observation samples are used to estimate its covariance matrix, which improves the filter's ability to track changes in speech signal and noise statistics.

In a specific embodiment of this application, we represent the time-domain noisy speech signal as,

y _m (t) = x _m (t) + v _m (t), m = 1,2,...,M (1)

Among them, y _m (t) represents the noisy speech signal received by the m-th microphone; x _m (t) represents the pure speech signal received by the m-th microphone; _vm (t) represents the m-th microphone received Background noise signal; t represents discrete time points; M represents the number of microphones.

In a specific embodiment of the present application, it is assumed that all signals are zero mean, bandwidth signals, and at the same time, it is assumed that the speech signal and the noise signal are not correlated. The purpose of speech noise reduction is to restore the pure speech signal through the noisy speech signal. Without loss of generality, in this application, the microphone 1 is set as the reference microphone, that is, x ₁ (t) is set as the desired signal (the signal to be restored).

The preprocessing of the noisy speech signal includes: framing the noisy speech signal, adding a window and then performing fast Fourier transform to obtain the frequency domain noisy speech signal, which is expressed as:

Among them, w represents the window function; T represents the length of the window function (also the length of the speech signal frame); L represents the step length between two adjacent frames; the zero-mean random variable Y _m (k,n), X _m (k,n), V _m (k, n) are y _m (t), x _m (t), v _m (t), respectively, the Fourier transform value of the k-th frequency band in the n-th frame, where k ∈{0,1,...,K-1}.

For convenience, the signal model is represented in vector form as

y(k,n)=x(k,n)+v(k,n) (3)

in,

y(k,n)=[Y ₁ (k,n),Y ₂ (k,n),...,Y _M (k,n)] ^T (4)

The definitions of x(k,n) and x(k,n) are similar to y(k,n), and the superscript T is the transpose operator.

In traditional methods, it is usually necessary to design a filter h(k,n) with a length of M to achieve speech noise reduction, namely:

Z(k,n)=h ^H (k,n)y(k,n) (5)

in

h(k,n)=[H ₁ (k,n),H ₂ (k,n),...,H _M (k,n)] ^T (6)

Z(k,n) is the estimated value of X ₁ (k,n). However, when M is large, two problems as described in the background art are caused.

Estimating the statistical properties of the noisy speech signal in the frequency domain, including estimating the statistical properties of the noisy speech signal according to the time smoothing method. Estimating the statistical properties of the noise signal, including estimating the statistical properties of the noise signal according to existing noise estimation algorithms.

Since the speech signal and noise are uncorrelated, the variance of Z(k,n) can be expressed as:

Φ _Z (k,n)=h ^H (k,n)Φ _y (k,n)h(k,n)

=h ^H (k,n)Φ _x (k,n)h(k,n)+h ^H (k,n)Φ _v (k,n)h(k,n) (7)

Among them, Φ _a (k,n)=E[a(k,n)a ^H (k,n)],a(k,n)∈{y(k,n),x(k,n),v (k,n)}. In general, we can estimate Φ _y (k,n) by applying temporal smoothing, and Φ _v (k,n) can be obtained according to noise estimation algorithms in the existing literature. After obtaining the estimated values of Φ _y (k, n) and Φ _v (k, n), Φ _x (k, n) can be obtained by Φ _y (k, n)-Φ _v (k, n).

In order to derive the method in the present invention, the microphone array is divided into M ₂ sub-arrays, and there are M ₁ microphones in each sub-array, that is, M=M ₁ *M ₂ , and the first to M ₁ microphones form the first sub-array , the M ₁ +1 to 2M ₁ microphones form the second sub-array, and so on. In the present invention, we assume that M ₁ ≤ M ₂ . Similarly, the filter h(k,n) can be decomposed as above, namely

in,

At this time, the sub-filters h _m (k, n), m=1, 2,..., M ₂ can be formed into a matrix of dimension M ₁ ×M ₂ , that is:

H(k,n)=[h ₁ (k,n),h ₂ (k,n),...,h _M2 (k,n)] (10)

It should be noted that h(k,n)=vec[H(k,n)], vec(·) represents the vectorization operator of the matrix. For brevity, the symbols k and n will be removed later where no ambiguity arises. Singular Value Decomposition (SVD) is performed on the matrix H, and H can be decomposed into:

in,

is an M ₂ ×M ₂ matrix,

上标H为共轭转置符。is an M ₂ ×M ₂ matrix. H ₁ and H ₂ are two orthogonal matrices, and Σ is an M ₁ ×M ₂ diagonal matrix whose diagonal elements are non-negative real numbers. In this application, they are arranged in descending order, i.e.

The superscript H is the conjugate transpose.

The noisy speech signals received by each channel are strongly correlated, so the sub-filters h _m (k, n), m=1, 2,..., M ₂ are usually also strongly correlated, resulting in the matrix H usually not Row full rank matrix. Therefore, the matrix H can usually be well approximated by the first P largest singular values and the corresponding singular vectors, namely:

需要注意的是，由

引起的歧义对矩阵H没有影响。相应的，滤波器h可以 近似表示为:in,

It should be noted that by

The resulting ambiguity has no effect on the matrix H. Correspondingly, the filter h can be approximately expressed as:

It should be noted that when P=M ₁ , h _P =h.

Apply the relation:

_hP can be written as:

大小为M×M₂，

大小为M×M₁。此时， 滤波器的输出值Z(k,n)可写为：in,

The size is M×M ₂ ,

The size is M×M ₁ . At this time, the output value Z(k,n) of the filter can be written as:

in,

H _σ1,P = [H _σ1,1 H _σ1,2 ...H _σ1,P ] ^H (24)

H _σ2,P = [H _σ2,1 H _σ2,2 ...H _σ2,P ] ^H (25)

The sizes of h _σ1,P , h _σ2,P , y _σ1,P (t), y _σ2,P (t), H _σ1,P and H _σ2,P are M ₁ P×1, M ₂ P×1, respectively , M ₂ P×1, M ₁ P×1, M ₂ P×M, M ₁ P×M. It can be seen that when the parameter P is small, the lengths of the sub-filters h _σ1,P and h _σ2,P are much smaller than the length of the filter h.

The mean square error (MSE) of the expected signal X ₁ and its estimated value Z is

E(·)表示数学期望，

表示取实部，上标^*表示复共轭。in,

E( ) represents the mathematical expectation,

Represents the real part, and the superscript ^* represents the complex conjugate.

For deriving the filter in the present invention, MSE is written in the following form:

in,

It should be noted that when the parameter P is small, the dimensions of the matrices Φ _yσ1,p (M ₂ P×M ₂ P), and Φ _yσ2,p (M ₁ P×M ₁ P) are much smaller than those of the matrix Φ _y ( M×M) dimension.

This leads to two advantages:

1) Compared with solving the traditional multi-channel speech noise reduction filter based on the inverse matrix of Φ _y , solving the sub-filters h _σ1,P and h _σ2,P based on the inverse matrix of Φ _yσ1,p and Φ _yσ2,p , The required complexity is significantly reduced;

2) Compared with the estimation matrix Φ _y , the matrices Φ _yσ1,p and Φ _yσ2,p can be estimated with fewer signal observation samples, so that the sub-filters h _σ1,P , and h _σ2,P can track the signal statistics more quickly changes in characteristics.

Calculating the approximate filter, including: adopting an iterative solution method to obtain a Wiener filter.

Based on equations (27) and (28), it is difficult to derive closed-form solutions for the subfilters h _σ1,P and h _σ2,P . Therefore, an iterative solution method is adopted in the present invention. For this reason, when solving one of the subfilters, the other subfilter is assumed to be fixed, i.e.

Initialize the subfilters h _{σ1, P} as follows:

in,

The definition of x _p is similar to that of y _p . It can be seen that h _σ1,W,p is the Vienna filter of the p-th sub-matrix, and its length is M ₁ .

构建

并将其带入式(29)和(30)，可得application

Construct

And bring it into equations (29) and (30), we can get

Substituting equations (38) and (39) into equation (34), we get:

求导并将结果置零，可得子滤波器

的维纳解:Put equation (40) on

Take the derivative and set the result to zero to get the subfilter

The Wiener solution of:

构建

并将其带入至式(31)和(32)，可得：application

Construct

and bringing it into equations (31) and (32), we get:

和

带入式(33)中得：Will

and

Bring it into equation (33) to get:

的维纳解：Based on (44), the sub-filter can be obtained

The Wiener solution of :

In the above way, when iterating to the nth step, we have:

in,

At this point, the iterative Wiener filter in this application can be obtained:

An embodiment of the present invention provides a voice noise reduction device based on a microphone array, as shown in FIG. 2, including a signal acquisition module 201, a signal preprocessing module 202, a statistical characteristic estimation module 203, a subfilter determination module 204, a frequency domain A noise reduction filter determination module 205 , a noise reduction module 206 , and a time domain noise reduction speech signal determination module 207 . The signal acquisition module 201 is used to acquire the noisy speech signal; the signal preprocessing module 202 is used to preprocess the noisy speech signal to determine the frequency domain noisy speech signal; the statistical characteristic estimation module 203 is used to estimate the The statistical characteristics of the frequency-domain noisy speech signal and the statistical characteristics of the noise signal are described; the sub-filter determination module 204 is used to divide the microphone array into a plurality of sub-arrays, and respectively estimate a plurality of sub-filters; the frequency-domain noise reduction filter determines Module 205, configured to determine a frequency-domain noise reduction filter according to the plurality of sub-filters; noise reduction module 206, configured to perform noise reduction processing on the frequency-domain noisy speech signal according to the frequency-domain noise reduction filter , determine the frequency-domain noise-reduced speech signal; the time-domain noise-reduced speech signal determining module 207 is configured to convert the frequency-domain noise-reduced speech signal into a time-domain noise-reduced speech signal.

Fig. 4 is a comparison between the complexity of the method provided by the application and the complexity of the traditional method, Fig. 5 is the change of the mean square error of the method provided by the application with the number of iterations, and Fig. 6 is the noise statistical characteristic suddenly changed, the application Plot of the mean squared error versus time for the proposed method and the traditional method. That is, the method provided by the present application effectively reduces the complexity and improves the ability of the filter to track changes in the statistical characteristics of speech signals and noise.

An embodiment of the present invention provides a voice noise reduction server based on a microphone array. As shown in FIG. 3 , it includes a memory 301 and a processor 302; the memory 301 is used to store computer-executable instructions; the processor 302 is used to execute computer-executable instructions. Execute the instruction.

Embodiments of the present invention provide a computer-readable storage medium, where the computer-readable storage medium stores executable instructions, and the computer can execute the executable instructions.

The above-mentioned storage medium includes but is not limited to random access memory (English: Random Access Memory; referred to as: RAM), read-only memory (English: Read-Only Memory; referred to as: ROM), cache (English: Cache), hard disk (English: Hard Disk Drive; referred to as: HDD) or memory card (English: Memory Card). The memory may be used to store computer program instructions.

Although the present application provides method operation steps as described in the embodiments or flow charts, more or less operation steps may be included based on routine or non-creative work. The sequence of steps enumerated in this embodiment is only one way among the execution sequences of many steps, and does not represent the only execution sequence. When an actual device or client product is executed, it can be executed sequentially or in parallel (for example, a parallel processor or multi-threaded processing environment) according to the method shown in this embodiment or the accompanying drawings.

The devices or modules described in the above embodiments may be specifically implemented by computer chips or entities, or by products with certain functions. For the convenience of description, when describing the above device, the functions are divided into various modules and described respectively. When implementing the present application, the functions of each module may be implemented in one or more software and/or hardware. Of course, a module that implements a certain function can also be implemented by a combination of multiple sub-modules or sub-units.

The methods, apparatuses or modules described in this application may be implemented in computer readable program code. The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and the memory may be implemented by the (micro)processing computer-readable medium, logic gates, switches, application-specific integrated circuits (English: Application Specific Integrated Circuit; ASIC for short), programmable logic controllers and embedded microcontrollers Examples of controllers include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320, and memory controllers can also be implemented as part of the memory's control logic. Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer-readable program code, the controller can be implemented as logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded devices by logically programming the method steps. The same function can be realized in the form of a microcontroller, etc. Therefore, such a controller can be regarded as a hardware component, and the devices included therein for realizing various functions can also be regarded as a structure within the hardware component. Or even, the means for implementing various functions can be regarded as both a software module implementing a method and a structure within a hardware component.

Some of the modules in the apparatus described herein may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products in essence, or the parts that make contributions to the prior art can also be embodied in the implementation process of data migration. The computer software product can be stored in a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes several instructions to make a computer device (which can be a personal computer, mobile terminal, server, or network device, etc.) execute this The methods described in various embodiments or portions of embodiments are claimed.

Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. All or part of this application may be utilized in numerous general purpose or special purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, including A distributed computing environment for any of the above systems or devices, and the like.

The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still The technical solutions are modified, or some or all of the technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the present application.

Claims (8)

1. A speech noise reduction method based on microphone array is characterized by comprising

Acquiring a voice signal with noise;

preprocessing the voice signal with noise to determine a frequency domain voice signal with noise;

estimating the statistical characteristic of the frequency domain voice signal with noise, and estimating the statistical characteristic of the noise signal;

dividing a microphone array into a plurality of sub-arrays, and respectively estimating a plurality of sub-filters;

determining a frequency domain noise reduction filter according to the plurality of sub-filters;

carrying out noise reduction processing on the frequency domain voice signal with noise according to the frequency domain noise reduction filter to determine a frequency domain noise reduction voice signal;

and converting the frequency domain noise reduction voice signal into a time domain noise reduction voice signal.

2. The method of claim 1, wherein the pre-processing the noisy speech signal comprises: and performing frame division and windowing on the voice signal with the noise, and then performing fast Fourier transform.

3. The method according to claim 1, wherein said estimating the statistical properties of the frequency-domain noisy speech signal comprises estimating the statistical properties of the noisy speech signal according to a time-smoothed estimation.

4. The method of claim 1, wherein estimating the statistical properties of the noise signal comprises estimating the statistical properties of the noise signal according to an existing noise estimation algorithm.

5. The method of claim 1, wherein the dividing the microphone array into a plurality of sub-arrays and estimating a plurality of sub-filters separately comprises iteratively estimating the plurality of sub-filters using a low rank architecture of a noise reduction filter.

6. A speech noise reduction device based on microphone array is characterized by comprising

The signal acquisition module is used for acquiring a voice signal with noise;

the signal preprocessing module is used for preprocessing the voice signal with the noise and determining a frequency domain voice signal with the noise;

the statistical characteristic estimation module is used for estimating the statistical characteristic of the frequency domain voice signal with noise and estimating the statistical characteristic of the noise signal;

the sub-filter determining module is used for dividing the microphone array into a plurality of sub-arrays and respectively estimating a plurality of sub-filters;

a frequency domain noise reduction filter determining module, configured to determine a frequency domain noise reduction filter according to the plurality of sub-filters;

the noise reduction module is used for carrying out noise reduction processing on the frequency domain voice signal with noise according to the frequency domain noise reduction filter and determining a frequency domain noise reduction voice signal;

and the time domain noise reduction voice signal determination module is used for converting the frequency domain noise reduction voice signal into a time domain noise reduction voice signal.

7. A microphone array based speech noise reduction server comprising a memory and a processor;

the memory is to store computer-executable instructions;

the processor is configured to execute the computer-executable instructions to implement the method of any of claims 1-5.

8. A computer-readable storage medium having stored thereon executable instructions that, when executed by a computer, are capable of implementing the method of any one of claims 1-5.

Images