Robust nonlinear echo cancellation method for visual intercom system

Abstract

The invention discloses a robust nonlinear echo cancellation method for a visual intercom system, which comprises the following steps: collecting a far-end voice signal of the visual intercom system as a reference signal to construct an input signal vector at the current moment, and carrying out nonlinear modeling on the reference signal through a random Fourier mapping method; acquiring a near-end signal of the visual intercom system to obtain a received expected signal; estimating an echo path based on an adaptive filtering algorithm, and obtaining an estimated echo signal through an input vector formed by a nonlinear extended reference signal; combining the estimated echo signal with the expected signal to eliminate echo, comparing the expected signal with the estimated echo signal, processing a difference signal through an adaptive control filter, and calculating a hyperbolic tangent distribution function; and updating the weight vector of the filter according to the hyperbolic tangent distribution function of the difference signal, and repeating the steps through iteration until the echo is eliminated. According to the invention, the audio communication quality of the visual intercom system is greatly improved.

Description

Technical Field

[0001] The invention belongs to the field of intercom systems, and in particular relates to a robust nonlinear echo elimination method for a visual intercom system. Background Art

[0002] Video intercom systems play a vital role in modern buildings and security monitoring, and are widely used in real-time communication scenarios. The system achieves efficient interaction between users by transmitting audio and video signals in both directions. However, since the speaker and microphone are usually integrated into the same device and the distance between them is small, the sound emitted by the speaker is easily captured by the microphone again, resulting in an echo phenomenon. The presence of echo seriously reduces the call quality and affects the user's communication experience. Therefore, echo cancellation technology has become the key to the development of video intercom systems.

[0003] Traditional echo cancellation technology mainly relies on linear filters, and its design concept is based on the linear assumption of the echo path. However, in actual application scenarios, due to the nonlinear distortion between the speaker and the microphone, this linear filter is difficult to completely eliminate the echo, especially in high volume or complex acoustic environments. With the improvement of audio quality requirements of video intercom systems and the need to adapt to complex acoustic environments, the existing linear echo cancellation technology has become insufficient. In recent years, with the advancement of nonlinear signal processing technology, researchers have proposed new echo cancellation methods based on nonlinear modeling. These new methods are more accurate in capturing the nonlinear relationship between the speaker and the microphone, thereby achieving better echo cancellation effects under complex acoustic conditions. Despite this, the existing nonlinear echo cancellation technology still faces the problems of high computational complexity and difficulty in real-time processing, and its robustness needs to be improved when the environmental noise changes significantly. Summary of the invention

[0004] In view of this, the present invention proposes an innovative echo cancellation method, which not only effectively deals with the echo problem caused by nonlinear distortion, but also maintains efficient echo suppression capability in a changeable and complex acoustic environment, while having good real-time processing performance and robustness to meet the strict requirements of the video intercom system for high-quality audio communication.

[0005] The present invention aims to provide a robust nonlinear echo cancellation method for a video intercom system to solve the problems of unsatisfactory echo cancellation effect, poor real-time performance and insufficient robustness in the prior art. The method can effectively eliminate echoes caused by nonlinear distortion in complex audio environments, ensuring that the system can maintain high-quality audio communication in different environments.

[0006] To achieve the above object, the present invention provides a robust nonlinear echo cancellation method for a video intercom system, comprising:

[0007] The far-end voice signal of the video intercom system is collected as a reference signal to construct an input signal vector at the current moment, and the reference signal is nonlinearly modeled by a random Fourier mapping method;

[0008] Collecting the near-end signal of the video intercom system, wherein the near-end signal includes an echo signal, a voice signal of a person in the room, and background noise, and obtaining a received expected signal;

[0009] The echo path is estimated based on an adaptive filtering algorithm, and an estimated echo signal is obtained by nonlinearly extending an input vector formed by a reference signal;

[0010] The estimated echo signal is combined with the expected signal to eliminate the echo, the expected signal is compared with the estimated echo signal, the difference signal is processed by an adaptive control filter, and a hyperbolic tangent distribution function is calculated; the weight vector of the filter is updated according to the hyperbolic tangent distribution function of the difference signal, and the above steps are repeated through iteration until the echo is eliminated.

[0011] Preferably, the process of collecting the far-end voice signal of the video intercom system as a reference signal to construct the input signal vector at the current moment includes:

[0012] Sampling is performed based on the outdoor unit of the remote device of the video intercom system to obtain the remote voice signal;

[0013] The far-end speech signal is used as the reference signal x i , and form the input signal vector x at the current sampling moment i , x i =x i ,x i-1 ,...,x i-L+1 T ; where L is the memory length, [g] T Represents the transpose operation.

[0014] Preferably, the process of performing nonlinear modeling on the reference signal by using a random Fourier mapping method comprises:

[0015] The reference signal is modeled using a random Fourier mapping method to generate a nonlinear extended reference signal V d,i Based on the nonlinear extended reference signal V d,i and the input signal vector to generate the L+D-dimensional input vector Ψ after mapping i .

[0016] Preferably, the reference signal is modeled using a random Fourier mapping method to generate a nonlinear extended reference signal V d,i The process includes:

[0017] Generate and select a set of random frequencies ωd,i and Phase For every random frequency ω d,i and Phase Construct the corresponding Fourier basis function; operate the reference signal with each Fourier basis function to obtain multiple nonlinear extended signal components V d,i , d=1,...,D;

[0018] Generate a nonlinear extended reference signal vector based on the nonlinear extended signal component obtained by random Fourier mapping The length of the vector after nonlinear expansion is D;

[0019] Each component of the nonlinearly extended reference signal is combined with the input signal vector to obtain a mapped input vector.

[0020] Preferably, the process of generating a random frequency includes:

[0021] At the current time i, generate a vector e with a mean of 0 and a covariance vector 2 I's Gaussian distribution generates a total of D groups, each with L frequencies of vector ω d,i ,ω d,i =[w1 w2 ... w L ],d=1,...,D;

[0022] The process of generating random phases includes:

[0023] Generate a phase with a uniform distribution on the interval [0,2π], with a total of D groups and 1 element in each group. d=1,...,D.

[0024] Preferably, the process of combining each component of the nonlinear extended reference signal with the input signal vector to obtain a mapped input vector comprises:

[0025] The input signal vector x i Each time point is taken as the 1:L column of the vector, and the nonlinear expansion reference signal vector Φ i Each component of is taken as the L+1:L+D column of the vector, and the augmented input vector is obtained

[0026] Preferably, the process of collecting the near-end signal of the video intercom system includes:

[0027] The far-end voice signal is transmitted to the room through the echo path to generate an echo signal r i , the echo signal and the indoor person's voice signal s i At the same time, it is collected by the near-end microphone to obtain the collected signal;

[0028] The collected signal and the indoor background noise v i Together they form the received desired signal d i , d i =r i +s i +v i .

[0029] Preferably, the process of estimating the echo path based on an adaptive filtering algorithm and obtaining the estimated echo signal through an input vector formed by nonlinearly extending the reference signal includes:

[0030] An adaptive filtering algorithm is used to estimate the echo path of the sound propagation change, and the input vector Ψ formed by the nonlinear expansion reference signal is i Through the adaptive filter C i Get the estimated echo signal y i ,

[0031] A negative signal is obtained according to the estimated echo signal, and the negative signal is combined with the mixed speech signal to eliminate the echo.

[0032] Preferably, the process of comparing the expected signal with the estimated echo signal, processing the difference signal by an adaptive control filter, and calculating the hyperbolic tangent distribution function comprises:

[0033] The expected signal d i and the estimated echo signal y i Compare and obtain a difference signal between the expected signal and the estimated echo signal;

[0034] The difference signal is input into the adaptive control filter for processing, and is updated to achieve the local minimum as the optimization target, and the hyperbolic tangent distribution function S of the difference signal at the current moment is calculated. i (e), S i (e i )=sech[ηe i ] ρ , where η is the scale parameter and ρ is the shape parameter.

[0035] Preferably, the process of updating the weight vector of the filter according to the hyperbolic tangent distribution function of the difference signal comprises:

[0036] Generate a gradient accumulation vector ▽S according to the hyperbolic tangent distribution function of the difference signal i , the gradient accumulation vector at the current time i

[0037] Based on the gradient descent method, according to the gradient accumulation vector ▽S i Update the weight vector C of the echo canceller at the next time i+1i+1 , C i+1 =C i -μ▽S i , where μ represents the update step size parameter.

[0038] Compared with the prior art, the present invention has the following advantages and technical effects:

[0039] The present invention proposes a robust nonlinear echo cancellation technology for a video intercom system. The technology converts a linear reference signal into a nonlinear form by means of random Fourier mapping, accurately matches the nonlinear distortion between the speaker and the microphone, and thus shows obvious superiority over traditional linear filters in dealing with nonlinear echo problems. At the same time, the present invention uses the hyperbolic tangent distribution function to construct a robustness optimization algorithm, ensuring that the system can maintain efficient echo cancellation performance under complex and changeable acoustic conditions, and protecting the call quality from environmental influences. Compared with the prior art, the present invention significantly improves the call quality, real-time processing capability and robustness of the video intercom system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0041] Figure 1 A schematic diagram of an echo cancellation system according to an embodiment of the present invention;

[0042] Figure 2 The figure is a schematic diagram of a method flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0043] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0044] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0045] like Figure 1-2 As shown, a robust nonlinear echo cancellation method for a video intercom system provided in this embodiment is an echo cancellation method based on nonlinear modeling and robustness optimization strategy. The control structure of the echo cancellation system is as follows: Figure 1 As shown, the steps are shown in S1-S7:

[0046] S1. Collect the remote voice signal from the outdoor unit of the video intercom system, use it as a reference signal and construct the input signal vector at the current moment;

[0047] S2, nonlinear modeling of the reference signal by random Fourier mapping method;

[0048] S3, collecting indoor near-end signals, including echo signals and indoor personnel voice signals, and background noise, to form received expected signals;

[0049] S4, adaptive estimation of the echo signal, using an adaptive filtering algorithm to estimate the echo path, and obtaining an estimated echo signal by nonlinearly extending an input vector formed by a reference signal, and then merging the estimated echo signal with the desired signal to eliminate the echo;

[0050] S5, acquiring and processing the error signal, comparing the expected signal with the estimated echo signal, processing the difference signal through an adaptive control filter, and calculating the hyperbolic tangent distribution function;

[0051] S6. updating the weight vector of the filter according to the hyperbolic tangent distribution function of the difference signal;

[0052] S7. Repeat the above steps iteratively until the echo is eliminated.

[0053] Further optimization solutions:

[0054] S1, outdoor remote input signal acquisition;

[0055] Based on the outdoor unit of the remote device of the video intercom system, the remote voice signal is obtained and the signal is used as the reference signal x i , and form the input signal vector x at the current sampling moment i , x i =x i ,x i-1 ,...,x i-L+1 T ; where L is the memory length, [g] T represents the transpose operation;

[0056] S2, nonlinear modeling of input signals;

[0057] The reference signal is modeled using the random Fourier mapping method to generate a nonlinear extended reference signal V d,i ; Generate the mapped L+D dimensional input vector Ψ according to the nonlinear extended reference signal and the input signal vector i .

[0058] The purpose of this step is to convert the linear reference signal into a nonlinear signal to match the nonlinear distortion.

[0059] The specific steps are as follows:

[0060] S21. Select a set of random frequencies ω d,i and Phase For each of the above random frequencies and phases, a corresponding Fourier basis function is constructed; the original reference signal is operated with each Fourier basis function to obtain multiple nonlinear extended signal components V d,i , d=1,...,D.

[0061] Furthermore, a random Fourier feature of the input signal is constructed; assuming is a continuous, symmetric and positively definite Mercer kernel. The Mercer kernel induces a nonlinear mapping F from the input space to the high-dimensional feature space H, satisfying the following equation where <·,g> H represents the inner product in space H. Then the output of the nonlinear system is expressed according to its representative theorem as where x i represents the input vector, a l ,l=1,2,L,k is the mapping coefficient for constructing nonlinear function.

[0062] Gaussian positive definite kernel function Where s>0 is the bandwidth parameter of the kernel function. According to the properties of the Gaussian function, the Gaussian kernel is transformed by Fourier transform to obtain:

[0063]

[0064] in represents Fourier transform, e 2 represents the variance. Then, p(ω) is inversely transformed by Fourier to obtain the expression of the kernel function: in Represents the inverse Fourier transform.

[0065] The above equation can be viewed as According to the definition of mathematical expectation, let The unbiased estimate of the kernel function can be obtained Based on the symmetry of the kernel, the real-valued mapping relationship of the nonlinear mapping F can be approximated as Where z(x) is the random Fourier characteristic of the input signal; ω d =[w1 w2 ... w L ],d=1,...,D follows the mean zero and the covariance vector e 2 Gaussian distribution of I. d=1,...,D obeys a uniform distribution on the interval [0,2π].

[0066] The acquisition of the above random Fourier features is given by S211-S213:

[0067] S211, generate random frequencies; at the current time i, generate a vector with a mean of 0 and a covariance vector e 2 I's Gaussian distribution generates a total of D groups, each with L frequencies of vector ω d,i ,ω d,i =[w1 w2...w L ],d=1,...,D;

[0068] S212, generate random phases; generate D groups of phases with 1 element in each group following a uniform distribution on the interval [0,2π] d=1,...,D;

[0069] S213, nonlinear extension of the reference signal; generating frequency and phase parameters according to steps S211 and S212 to obtain a nonlinear extended reference signal d=1,...,D.

[0070] S22, generating a nonlinear extended reference signal vector according to the nonlinear extended signal component obtained by the above random Fourier mapping The length of the vector after nonlinear expansion is D.

[0071] S23, combining each component of the nonlinear extended reference signal with the input signal vector to form a mapped input vector. The vector is constructed as follows:

[0072] The above input signal vector x i Each time point of is a 1:L column of the vector;

[0073] The above nonlinear extended reference signal Φ i Each component of is taken as the L+1:L+D column of the vector;

[0074] Through the above combination, we get an augmented input vector Ψ i , used for subsequent adaptive filtering processing.

[0075] S3, indoor near-end signal acquisition;

[0076] The far-end voice signal is transmitted to the room through the echo path to generate an echo signal r i , echo signal and indoor voice signal s i At the same time, it is collected by the near-end microphone; the above collected signal and the indoor background noise v i Together they form the received desired signal d i , d i =ri +s i +v i .

[0077] S4, adaptive estimation of echo signal;

[0078] Adaptive filtering algorithm is used to estimate the echo path of sound propagation changes. The nonlinear extended reference signal V i The input vector Ψ is formed i Through the adaptive filter C i Get the estimated echo signal y i , A negative signal is obtained according to the estimated echo signal, and the negative signal is combined with the mixed speech signal to eliminate the echo.

[0079] S5, error signal acquisition and processing;

[0080] From the above expected signal d i With the estimated echo signal y i The difference signal between the expected signal and the estimated echo signal is compared to obtain the difference signal; the difference signal is input into the adaptive control filter for processing, and is updated to achieve the local minimum as the optimization goal, and the hyperbolic tangent distribution function S of the difference signal at the current moment is calculated. i (e), S i (e i )=sech[ηe i ] ρ , where η is the scale parameter and ρ is the shape parameter;

[0081] S6, filter update;

[0082] Generate the gradient accumulation vector ▽S according to the hyperbolic tangent distribution function of the difference signal i , the gradient accumulation vector at the current time i

[0083] Based on the gradient descent method, the weight vector C of the echo canceller at the next time i+1 is updated according to the gradient accumulation vector i+1 , C i+1 =C i -μ▽S i , where μ represents the update step size parameter.

[0084] S7. Iteratively repeat steps S1 to S7 until the echo cancellation is completed.

[0085] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A robust nonlinear echo cancellation method for a video intercom system, characterized in that: include: The far-end voice signal of the video intercom system is collected as a reference signal to construct an input signal vector at the current moment, and the reference signal is nonlinearly modeled by a random Fourier mapping method; Collecting the near-end signal of the video intercom system, wherein the near-end signal includes an echo signal, a voice signal of a person in the room, and background noise, and obtaining a received expected signal; The echo path is estimated based on an adaptive filtering algorithm, and an estimated echo signal is obtained by nonlinearly extending an input vector formed by a reference signal; The estimated echo signal is combined with the expected signal to eliminate the echo, the expected signal is compared with the estimated echo signal, the difference signal is processed by an adaptive control filter, and a hyperbolic tangent distribution function is calculated; the weight vector of the filter is updated according to the hyperbolic tangent distribution function of the difference signal, and the above steps are repeated through iteration until the echo is eliminated.

2. The robust nonlinear echo cancellation method for a video intercom system according to claim 1, characterized in that: The process of collecting the far-end voice signal of the video intercom system as a reference signal to construct the input signal vector at the current moment includes: Sampling is performed based on the outdoor unit of the remote device of the video intercom system to obtain the remote voice signal; The far-end speech signal is used as the reference signal x i , and form the input signal vector x at the current sampling moment i , x i =x i ,x i-1 ,...,x i-L+1 T ; where L is the memory length, [g] T Represents the transpose operation.

3. The robust nonlinear echo cancellation method for a video intercom system according to claim 1, characterized in that: The process of performing nonlinear modeling on the reference signal by using a random Fourier mapping method includes: The reference signal is modeled using a random Fourier mapping method to generate a nonlinear extended reference signal Based on the nonlinear extended reference signal V d,i and the input signal vector to generate the L+D-dimensional input vector Ψ after mapping i .

4. The robust nonlinear echo cancellation method for a video intercom system according to claim 3, characterized in that: The reference signal is modeled using a random Fourier mapping method to generate a nonlinear extended reference signal V d,i The process includes: Generate and select a set of random frequencies ω d,i and Phase For every random frequency ω d,i and Phase Constructing corresponding Fourier basis functions; operating the reference signal with each Fourier basis function to obtain multiple nonlinear extended signal components Generate a nonlinear extended reference signal vector based on the nonlinear extended signal component obtained by random Fourier mapping The length of the vector after nonlinear expansion is D; Each component of the nonlinear extended reference signal is combined with the input signal vector to obtain a mapped input vector.

5. The robust nonlinear echo cancellation method for a video intercom system according to claim 4, characterized in that: The process of generating random frequencies includes: At the current time i, generate a vector e with a mean of 0 and a covariance vector 2 I's Gaussian distribution generates a total of D groups, each with L frequencies of vector ω d,i ,ω d,i =[w1 w2...w L ],d=1,...,D; The process of generating random phases includes: Generate a phase with a uniform distribution on the interval [0,2π], with a total of D groups and 1 element in each group.

6. The robust nonlinear echo cancellation method for a video intercom system according to claim 4, characterized in that: The process of combining the components of the nonlinear extended reference signal with the input signal vector to obtain the mapped input vector includes: The input signal vector x i Each time point is taken as the 1:L column of the vector, and the nonlinear expansion reference signal vector Φ i Each component of is taken as the L+1:L+D column of the vector, and the augmented input vector is obtained 7. The robust nonlinear echo cancellation method for a video intercom system according to claim 1, characterized in that: The process of collecting the near-end signal of the video intercom system includes: The far-end voice signal is transmitted to the room through the echo path to generate an echo signal r i , the echo signal and the indoor person's voice signal s i At the same time, it is collected by the near-end microphone to obtain the collected signal; The collected signal and the indoor background noise v i Together they form the received desired signal d i , d i =r i +s i +v i .

8. The robust nonlinear echo cancellation method for a video intercom system according to claim 1, characterized in that: The process of estimating the echo path based on the adaptive filtering algorithm and obtaining the estimated echo signal through the input vector formed by the nonlinear expansion reference signal includes: An adaptive filtering algorithm is used to estimate the echo path of the sound propagation change, and the input vector Ψ formed by the nonlinear expansion reference signal is i Through the adaptive filter C i Get the estimated echo signal y i ,y i =C i T Ψ i ; A negative signal is obtained according to the estimated echo signal, and the negative signal is combined with the mixed speech signal to eliminate the echo.

9. The robust nonlinear echo cancellation method for a video intercom system according to claim 1, characterized in that: The process of comparing the expected signal with the estimated echo signal, processing the difference signal through an adaptive control filter, and calculating the hyperbolic tangent distribution function includes: The expected signal d i and the estimated echo signal y i Compare and obtain a difference signal between the expected signal and the estimated echo signal; The difference signal is input into the adaptive control filter for processing, and is updated to achieve the local minimum as the optimization target, and the hyperbolic tangent distribution function S of the difference signal at the current moment is calculated. i (e), S i (e i )=sech[ηe i ] ρ , where η is the scale parameter and ρ is the shape parameter.

10. The robust nonlinear echo cancellation method for a video intercom system according to claim 1, characterized in that: The process of updating the weight vector of the filter according to the hyperbolic tangent distribution function of the difference signal includes: Generate a gradient accumulation vector according to the hyperbolic tangent distribution function of the difference signal The gradient accumulation vector at the current time i Based on the gradient descent method, according to the gradient accumulation vector Update the weight vector C of the echo canceller at the next time i+1 i+1 , Where μ represents the update step size parameter.

Images