CN102740214B

CN102740214B - Howling suppression method based on feedback signal spectrum estimation

Info

Publication number: CN102740214B
Application number: CN201110082197.4A
Authority: CN
Inventors: 杨飞然; 吴鸣; 杨军
Original assignee: Institute of Acoustics CAS
Current assignee: In New Science And Technology Co Ltd (suzhou)
Priority date: 2011-04-01
Filing date: 2011-04-01
Publication date: 2014-09-10
Anticipated expiration: 2031-04-01
Also published as: CN102740214A

Abstract

The present invention relates to a howling suppression method based on feedback signal spectrum estimation. Firstly, the signal collected by the microphone and the output signal of the loudspeaker are divided into frames, windowed and Fourier transformed to obtain the corresponding frequency domain signal; then, according to the obtained The microphone frequency domain signal and the loudspeaker output frequency domain signal calculate the cross power spectrum between the microphone and the loudspeaker output signal and the power spectrum of the loudspeaker output signal; and divide the cross power spectrum by the power spectrum to obtain an estimate of the magnitude spectrum of the feedback path transfer function; Then use the obtained feedback path transfer function magnitude spectrum estimate to calculate the current frame feedback signal magnitude spectrum; then use the microphone signal magnitude spectrum and the obtained feedback signal magnitude spectrum to calculate the gain function; finally, according to the obtained gain function, the microphone frequency domain The signal is corrected to achieve howling suppression. This method has fast convergence and tracking speed, good robustness, and low algorithm complexity, which is conducive to real-time implementation.

Description

A Howling Suppression Method Based on Feedback Signal Spectrum Estimation

技术领域 technical field

本发明涉及啸叫抑制领域，特别涉及一种基于反馈信号频谱估计的啸叫抑制方法。The invention relates to the field of howling suppression, in particular to a howling suppression method based on feedback signal frequency spectrum estimation.

背景技术 Background technique

啸叫抑制在有的文献也被称为“声反馈抑制”、“声反馈控制”、“反馈消除”等，它们表达的是同一种意思。如图1所示，图1为啸叫产生原理图。s(n)表示近端语音，这是需要放大的信号；d(n)表示从扬声器发出的声音经过反馈路径102再次反馈到麦克风，被麦克风拾取到的反馈信号。d(n)不是期望接收的信号，需要把它抑制掉。前向处理模块104用来完成啸叫抑制和自动增益控制等功能。输入信号s(n)到输出信号y(n)之间的传递函数为G(z)/(1-G(z)F(z))，如果同时满足(a)|G(z)F(z)|≥1；(b)∠G(z)F(z)＝2πn，n∈N；则该系统将变得不稳定，在满足条件的频率点上就会发生啸叫。使用扩音系统目的之一就是为了提高系统增益，助听器的佩戴者也希望能够把声音放大到所期望的水平，啸叫现象限制了扩声系统的增益量。因此，必须采取一定的措施消除反馈来避免啸叫的发生。公共扩声系统中由于讲话人的位置可能会经常地变化。例如：教室里老师授课用的扩音系统中，麦克风被放置在老师的衣服上而老师在授课时身体的位置不是固定的，这就导致反馈路径的不断变化，相应的解决方案必须考虑到这一点并能快速的跟踪这种变化，否则就会出现啸叫。Howling suppression is also called "acoustic feedback suppression", "acoustic feedback control", "feedback elimination" and so on in some literatures, and they express the same meaning. As shown in Figure 1, Figure 1 is a schematic diagram of howling generation. s(n) represents the near-end voice, which is a signal that needs to be amplified; d(n) represents the sound emitted from the speaker and fed back to the microphone through the feedback path 102, and the feedback signal picked up by the microphone. d(n) is not the desired signal and needs to be suppressed. The forward processing module 104 is used to complete functions such as howling suppression and automatic gain control. The transfer function between the input signal s(n) and the output signal y(n) is G(z)/(1-G(z)F(z)), if (a)|G(z)F( z)|≥1; (b)∠G(z)F(z)=2πn, n∈N; then the system will become unstable, and howling will occur at the frequency points that meet the conditions. One of the purposes of using the sound reinforcement system is to increase the system gain. The wearer of the hearing aid also hopes to amplify the sound to the desired level. The howling phenomenon limits the gain of the sound reinforcement system. Therefore, certain measures must be taken to eliminate feedback to avoid the occurrence of howling. In a public sound reinforcement system, the speaker's position may change frequently. For example: in the PA system used by the teacher to teach in the classroom, the microphone is placed on the teacher's clothes and the position of the teacher's body is not fixed when teaching, which leads to constant changes in the feedback path, and the corresponding solution must take this into account One point and can quickly track this change, otherwise there will be howling.

啸叫抑制是一个热门的研究课题。目前的解决方法主要有：Howling suppression is a hot research topic. The current solutions mainly include:

(1)移频法。移频法通过破坏啸叫产生的相位条件来达到啸叫抑制的目的。然而，这种方法的啸叫抑制效果并不好，常常控制了一个频率点的啸叫，系统又会在别的频率点产生啸叫。而且移频法对语音质量有可觉察的影响，研究和实际系统测试表明使用移频法最多可以获得6dB的增益。(1) Frequency shift method. The frequency shifting method achieves the purpose of howling suppression by destroying the phase condition of howling. However, the howling suppression effect of this method is not good, and often the howling at one frequency point is controlled, and the system will generate howling at other frequency points. Moreover, the frequency shifting method has a perceptible impact on the voice quality. Research and actual system tests show that the frequency shifting method can obtain a gain of up to 6dB.

(2)陷波器法。陷波器法首先寻找可能发生或者已经发生了反馈的频谱位置，然后将信号中这些过强的频率成分衰减掉来达到反馈抑制效果。但是在实际系统中，有可能存在多个反馈频率点。因此，陷波器法效果有限。已经报道的使用陷波器法可以提供3dB-8dB的增益。(2) Notch filter method. The notch filter method first looks for the frequency spectrum position where feedback may occur or has occurred, and then attenuates these excessive frequency components in the signal to achieve the feedback suppression effect. However, in an actual system, there may be multiple feedback frequency points. Therefore, the effect of the notch filter method is limited. It has been reported that using the notch filter method can provide a gain of 3dB-8dB.

(3)自适应滤波器法。自适应滤波器法的工作原理是利用滤波器来模拟声音通过扬声器发出后再从麦克风进来这样的一个传播通道的传输特性，使得从滤波器出来的信号和麦克风采集到的反馈信号一致，然后从麦克风的输入信号中把该部分信号减掉来达到消除反馈的目的。如图2所示，图2为基于自适应滤波算法的啸叫抑制系统框图。自适应滤波单元206的参数必须调整到和实际传输路径特性接近，其模拟得到的信号便越接近实际的反馈信号，就可以取得更好的啸叫抑制效果。这种方法和回声抵消所采用的方法原理是一样的，但是与回声抵消不同的是在反馈场合下，系统总是处于被称为“双端对讲”的状态，从而影响了自适应滤波器的收敛速度，而且估计出的滤波器系数是有偏的，这种方法很难适用于反馈路径总是在不断变化的情况。(3) Adaptive filter method. The working principle of the adaptive filter method is to use the filter to simulate the transmission characteristics of a propagation channel where the sound is sent out through the speaker and then comes in from the microphone, so that the signal from the filter is consistent with the feedback signal collected by the microphone, and then from the This part of the signal is subtracted from the input signal of the microphone to achieve the purpose of eliminating feedback. As shown in Figure 2, Figure 2 is a block diagram of a howling suppression system based on an adaptive filtering algorithm. The parameters of the adaptive filtering unit 206 must be adjusted to be close to the characteristics of the actual transmission path, and the closer the simulated signal is to the actual feedback signal, the better the howling suppression effect can be achieved. This method is the same as the method used in echo cancellation, but the difference from echo cancellation is that in the case of feedback, the system is always in a state called "double-ended intercom", which affects the adaptive filter. The convergence speed, and the estimated filter coefficients are biased, this method is difficult to apply to the situation where the feedback path is always changing.

另外，自适应滤波器法分为不连续啸叫抑制和连续啸叫抑制。不连续啸叫抑制的主要缺点是这种方法需要中断正常输入语音信号，从而使语音的可懂度受到影响，因而在很多场合是不能被接受的。而连续啸叫抑制使用连续的输入信号来估计反馈信号，但是收敛速度很慢。In addition, the adaptive filter method is divided into discontinuous howling suppression and continuous howling suppression. The main disadvantage of discontinuous howling suppression is that this method needs to interrupt the normal input voice signal, thus affecting the intelligibility of the voice, so it is unacceptable in many occasions. While continuous howling suppression uses continuous input signals to estimate the feedback signal, but the convergence speed is very slow.

综上所述，目前的啸叫抑制方法鲁棒性不好，当反馈路径变化时都很可能出现短暂的啸叫。To sum up, the current howling suppression method is not robust, and transient howling is likely to occur when the feedback path changes.

发明内容 Contents of the invention

本发明的目的在于，为了解决上述问题，提出一种基于反馈信号频谱估计的啸叫抑制方法，从而达到鲁棒性好，收敛和跟踪速度快的目的。The object of the present invention is to propose a howling suppression method based on feedback signal spectrum estimation to solve the above problems, so as to achieve the goals of good robustness, fast convergence and tracking speed.

为实现上述发明目的，本发明提出一种基于反馈信号频谱估计的啸叫抑制方法，该方法具体步骤包括：In order to achieve the purpose of the above invention, the present invention proposes a howling suppression method based on feedback signal spectrum estimation. The specific steps of the method include:

步骤1)：分别对麦克风采集的信号x(n)和扬声器输出信号y(n)进行分帧、加窗和傅里叶变换得到麦克风频域信号X(i，k)和扬声器输出频域信号Y(i，k)；Step 1): Framing, windowing, and Fourier transform are performed on the signal x(n) collected by the microphone and the output signal y(n) of the speaker to obtain the frequency domain signal X(i, k) of the microphone and the output frequency domain signal of the speaker Y(i,k);

步骤2)：根据所述的步骤1)获得的麦克风频域信号X(i，k)和扬声器输出频域信号Y(i，k)计算麦克风和扬声器输出信号之间的互功率谱φ_yx(i，k)和扬声器输出信号的功率谱φ_yy(i，k)；并将互功率谱φ_yx(i，k)除以功率谱φ_yy(i，k)得到反馈路径传递函数幅度谱的估计 Step 2): According to the microphone frequency domain signal X (i, k) obtained in the step 1) and the loudspeaker output frequency domain signal Y (i, k) calculate the cross power spectrum φ _yx ( i, k) and the power spectrum φ _yy (i, k) of the speaker output signal; and dividing the cross power spectrum φ _yx (i, k) by the power spectrum φ _yy (i, k) to obtain the feedback path transfer function magnitude spectrum estimate

步骤3)：利用所述的步骤2)获得的反馈路径传递函数幅度谱的估计计算得到当前帧反馈信号幅度谱 Step 3): Using the feedback path transfer function amplitude spectrum obtained in step 2) to estimate Calculate the magnitude spectrum of the current frame feedback signal

步骤4)：利用麦克风信号幅度谱|X(i，k)|和所述的步骤3)获得的反馈信号幅度谱计算获得增益函数G(i，k)；Step 4): Using the microphone signal magnitude spectrum |X(i, k)| and the feedback signal magnitude spectrum obtained in step 3) Calculate and obtain the gain function G(i, k);

步骤5)：根据所述的步骤4)获得的增益函数G(i，k)来对麦克风频域信号X(i，k)进行修正来实现啸叫抑制。Step 5): According to the gain function G(i, k) obtained in the above step 4), the microphone frequency domain signal X(i, k) is modified to realize howling suppression.

该方法还进一步包括：The method further includes:

步骤6)：对所述的步骤5)获得的修正后的麦克风频域信号进行处理获得最终的输出信号并对输出信号进行解相关处理；和Step 6): Process the corrected microphone frequency domain signal obtained in step 5) to obtain the final output signal and for the output signal performing a decorrelation process; and

步骤7)：对所述的步骤6)处理的结果进行增益调节。Step 7): Perform gain adjustment on the processing result of step 6).

所述的步骤2)获得反馈路径传递函数幅度谱的具体步骤包括：Described step 2) the specific steps of obtaining the magnitude spectrum of the feedback path transfer function include:

21)：采用一阶自回归平滑的方法，按照式(1)获得互功率谱φ_yx(i，k)；21): Using the method of first-order autoregressive smoothing, the cross power spectrum φ _yx (i, k) is obtained according to formula (1);

φ_yx(i，k)＝α_Cφ_yx(i-1，k)+(1-α_C)|Y^*(i，k)X(i，k)| (1)φ _yx (i, k)=α _C φ _yx (i-1, k)+(1-α _C )|Y ^* (i, k)X(i, k)| (1)

其中，α_C是平滑因子，0≤α_C＜1；Among them, α _C is a smoothing factor, 0≤α _C <1;

22)：采用一阶自回归平滑的方法，按照式(2)获得功率谱φ_yy(i，k)；22): Using the first-order autoregressive smoothing method, the power spectrum φ _yy (i, k) is obtained according to formula (2);

φ_yy(i，k)＝α_Rφ_yy(i-1，k)+(1-α_R)Y^*(i，k)Y(i，k) (2)φ _yy (i, k) = α _R φ _yy (i-1, k) + (1-α _R ) Y ^* (i, k) Y(i, k) (2)

其中，α_R是平滑因子，0≤α_R＜1；Among them, α _R is a smoothing factor, 0≤α _R <1;

23)：利用互功率谱φ_yx(i，k)和功率谱φ_yy(i，k)根据式(3)获得反馈路径传递函数幅度谱 23): Use cross power spectrum φ _yx (i, k) and power spectrum φ _yy (i, k) to obtain feedback path transfer function magnitude spectrum according to formula (3)

$| | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | = = \frac{{φ φ}_{yx yx} ((i i,, k k))}{{φ φ}_{yy yy} ((i i,, k k)) + + δ δ} - - - - - - ((33))$

其中，δ是一个很小的正数来避免式(3)的分母为0。Among them, δ is a very small positive number to avoid the denominator of formula (3) being 0.

所述的步骤3)按照式(4)计算获得当前帧反馈信号幅度谱 The step 3) is calculated according to formula (4) to obtain the current frame feedback signal amplitude spectrum

$| | \overset{^^}{X x} ((i i,, k k)) | | = = β β | | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | | | X x ((i i,, k k)) | | - - - - - - ((44))$

其中，β是一个可调的参数，根据实际情况进行合理选择该参数。Among them, β is an adjustable parameter, which should be reasonably selected according to the actual situation.

所述的步骤4)计算增益函数的方法包括：基于先验信号反馈比方法和基于后验信号反馈比方法。The method for calculating the gain function in step 4) includes: a method based on a priori signal feedback ratio and a method based on a posteriori signal feedback ratio.

所述的步骤4)基于后验信号反馈比方法按照式(5)计算获得增益函数G(i，k)；The step 4) calculates and obtains the gain function G(i, k) based on the posterior signal feedback ratio method according to formula (5);

$G G ((i i,, k k)) = = {[[\frac{max max (({| | X x ((i i,, k k)) | |}^{α α} - - {| | \overset{^^}{X x} ((i i,, k k)) | |}^{α α},, 00))}{{| | X x ((i i,, k k)) | |}^{α α}}]]}^{\frac{11}{α α}} - - - - - - ((55))$

其中，α＞0。为减小计算复杂度，α取1或2。Wherein, α>0. In order to reduce the computational complexity, α takes 1 or 2.

所述的步骤5)按照式(6)对麦克风频域信号X(i，k)进行修正来实现啸叫抑制；Step 5) modifying the microphone frequency domain signal X(i, k) according to formula (6) to realize howling suppression;

$\overset{^^}{S S} ((i i,, k k)) = = X x ((i i,, k k)) G G ((i i,, k k)) - - - - - - ((66)) . .$

所述的步骤7)采用调节数字信号的增益或调节模拟系统功放的增益。本发明的优点在于，与现有技术相比，本发明具有很好的鲁棒性；初始收敛速度快而且在反馈路径变化时能迅速跟踪这种变化而不会发生啸叫；由于本发明的所有操作都是频域进行的，因此具有很低的计算复杂度。The step 7) adopts adjusting the gain of the digital signal or adjusting the gain of the power amplifier of the analog system. The advantages of the present invention are that, compared with the prior art, the present invention has good robustness; the initial convergence speed is fast and can quickly track this change without howling when the feedback path changes; due to the All operations are performed in the frequency domain and thus have very low computational complexity.

附图说明 Description of drawings

图1为啸叫产生原理图；Figure 1 is a schematic diagram of howling generation;

图2为传统的基于自适应滤波算法的啸叫抑制系统框图；Figure 2 is a block diagram of a traditional howling suppression system based on an adaptive filtering algorithm;

图3为本发明的反馈路径估计系统框图；Fig. 3 is a block diagram of the feedback path estimation system of the present invention;

图4为本发明的基于反馈信号频谱估计的啸叫抑制方法结构框图；FIG. 4 is a structural block diagram of a howling suppression method based on feedback signal spectrum estimation in the present invention;

图5为本发明的反馈频谱估计模块410的框图；FIG. 5 is a block diagram of the feedback spectrum estimation module 410 of the present invention;

图6为实测的一个反馈路径脉冲响应图；Figure 6 is a measured feedback path impulse response diagram;

图7为实际反馈路径传递函数的幅度谱和本发明估计出来的反馈路径传递函数的幅度谱对比图。Fig. 7 is a comparison diagram of the magnitude spectrum of the actual feedback path transfer function and the magnitude spectrum of the feedback path transfer function estimated by the present invention.

具体实施方式 Detailed ways

下面结合附图和具体实施例对本发明进行详细的说明。The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

本发明的目的是提出了一种鲁棒性好，收敛和跟踪速度快，在有效抑制啸叫的同时保证语音质量不受损失的啸叫抑制方法。The object of the present invention is to propose a howling suppression method with good robustness, fast convergence and tracking speed, and effective suppression of howling while ensuring that voice quality is not lost.

为了达到上述目的，本发明采取的技术方案如下：In order to achieve the above object, the technical scheme that the present invention takes is as follows:

一种基于反馈信号频谱估计的啸叫抑制方法，具体步骤包括：A howling suppression method based on feedback signal spectrum estimation, the specific steps include:

步骤1)：分别对采集到的麦克风信号和扬声器输出信号进行加窗处理和傅里叶变换，将时域信号变换到频域；Step 1): Perform windowing processing and Fourier transform on the collected microphone signal and loudspeaker output signal respectively, and transform the time domain signal into the frequency domain;

步骤2)：利用平滑的麦克风信号和扬声器输出信号的互功率谱除以平滑的扬声器输出信号的功率谱得到反馈路径的幅度谱，并利用反馈路径的幅度谱计算出当前帧反馈信号的幅度谱；Step 2): divide the cross power spectrum of the smooth microphone signal and the speaker output signal by the power spectrum of the smooth speaker output signal to obtain the magnitude spectrum of the feedback path, and use the magnitude spectrum of the feedback path to calculate the magnitude spectrum of the current frame feedback signal ;

步骤3)：利用麦克风信号幅度谱和反馈信号幅度谱计算增益函数；Step 3): using the microphone signal amplitude spectrum and the feedback signal amplitude spectrum to calculate the gain function;

步骤4)：对麦克风信号频谱进行修正；Step 4): correcting the spectrum of the microphone signal;

步骤5)：对修正后的麦克风信号频谱进行逆傅里叶变换得到时域输出；Step 5): Inverse Fourier transform is performed on the corrected microphone signal spectrum to obtain a time domain output;

步骤6)：可选的)对步骤5)的输出施加一个解相关模块；Step 6): optional) applying a decorrelation module to the output of step 5);

步骤7)：可选的)对步骤6)的输出施加一个增益。Step 7): Optional) Apply a gain to the output of step 6).

在上述技术方案中，进一步地，步骤1)中所述采用帧处理的模式进行信号处理，所有处理和程序都是基于帧为单位的。需要对接收到的一帧麦克风信号和扬声器的输出信号进行加窗处理，然后使用FFT变换到频域。In the above technical solution, further, in step 1), the frame processing mode is used for signal processing, and all processing and procedures are based on frame units. It is necessary to perform windowing processing on the received frame of the microphone signal and the output signal of the speaker, and then use FFT to transform it into the frequency domain.

在上述技术方案中，进一步地，步骤2)中所述首先需要估计反馈路径传递函数的幅度谱，并利用它来进一步计算当前帧反馈信号的幅度谱。反馈路径传递函数的幅度谱是利用平滑的麦克风信号和扬声器输出信号的互功率谱绝对值除以平滑的扬声器输出信号的功率谱而得到的。In the above technical solution, further, in step 2), it is first necessary to estimate the magnitude spectrum of the transfer function of the feedback path, and use it to further calculate the magnitude spectrum of the feedback signal of the current frame. The magnitude spectrum of the feedback path transfer function is obtained by dividing the absolute value of the cross power spectrum of the smoothed microphone signal and the loudspeaker output signal by the power spectrum of the smoothed loudspeaker output signal.

在上述的技术方案中，进一步地，步骤3)所述增益函数是基于后验信号反馈比计算得到的。In the above technical solution, further, the gain function in step 3) is calculated based on the posterior signal feedback ratio.

在上述的技术方案中，进一步地，步骤4)对麦克风的频谱进行修正是指仅对其幅度谱进行修正，而保持其相位谱不变。In the above technical solution, further, step 4) modifying the frequency spectrum of the microphone refers to only modifying its amplitude spectrum while keeping its phase spectrum unchanged.

在上述的技术方案中，进一步，步骤5)是对修正后的麦克风频谱进行逆傅里叶变换来得到时域信号，然后与前一帧的输出信号进行重叠叠加作为啸叫抑制的输出信号。In the above technical solution, further, step 5) is to inverse Fourier transform the corrected microphone spectrum to obtain a time-domain signal, and then overlap and superimpose it with the output signal of the previous frame as the howling suppression output signal.

在上述的技术方案中，进一步，步骤6)所述的解相关模块用来去除扬声器输出信号和麦克风接收到的近端信号之间的相关性。In the above technical solution, further, the de-correlation module described in step 6) is used to remove the correlation between the speaker output signal and the near-end signal received by the microphone.

在上述的技术方案中，进一步，步骤7)所述的增益既可以加在数字系统的数字信号上也可以加在模拟系统的功放端。In the above technical solution, further, the gain described in step 7) can be added to the digital signal of the digital system or to the power amplifier end of the analog system.

如图4所示，图4为本发明的基于反馈信号频谱估计的啸叫抑制方法结构框图。其中，s(n)表示近端信号，d(n)表示反馈信号，x(n)表示麦克风采集到的信号，y(n)表示扬声器输出的信号。As shown in FIG. 4 , FIG. 4 is a structural block diagram of a howling suppression method based on feedback signal spectrum estimation in the present invention. Wherein, s(n) represents a near-end signal, d(n) represents a feedback signal, x(n) represents a signal collected by a microphone, and y(n) represents a signal output by a speaker.

步骤1)：分别对采集到的麦克风信号和扬声器输出信号进行加窗处理和傅里叶变换，将时域信号变换到频域。Step 1): Perform windowing processing and Fourier transform on the collected microphone signal and speaker output signal respectively, and transform the time domain signal into the frequency domain.

对采集到的麦克风信号x(n)进行串并变换后经过加窗模块402和傅里叶变换模块404后将时域信号变换到频域；对扬声器输出信号y(n)进行串并变换后经过加窗模块402和傅里叶变换模块404后将时域信号变换到频域。y(n)的第i帧离散傅里叶变换记为Y(i，k)，x(n)的第i帧离散傅里叶变换记为X(i，k)。After the serial-parallel conversion of the collected microphone signal x(n), the time-domain signal is converted to the frequency domain after passing through the windowing module 402 and the Fourier transform module 404; after the serial-parallel conversion of the speaker output signal y(n) After the windowing module 402 and the Fourier transform module 404, the time domain signal is transformed into the frequency domain. The discrete Fourier transform of the i-th frame of y(n) is denoted as Y(i, k), and the discrete Fourier transform of the i-th frame of x(n) is denoted as X(i, k).

步骤2)：利用平滑的麦克风信号x(n)和扬声器输出信号y(n)的互功率谱除以平滑的扬声器输出信号y(n)的功率谱得到反馈路径的幅度谱，并利用反馈路径的幅度谱计算出当前帧反馈信号的幅度谱。Step 2): Divide the cross power spectrum of the smoothed microphone signal x(n) and the speaker output signal y(n) by the power spectrum of the smoothed speaker output signal y(n) to obtain the magnitude spectrum of the feedback path, and use the feedback path The amplitude spectrum of the current frame is calculated to calculate the amplitude spectrum of the feedback signal.

如图3所示，图3为本发明的反馈路径估计系统框图。麦克风拾取的信号为x(n)＝d(n)+s(n)，d(n)是扬声器输出信号y(n)通过反馈路径传递函数F(z)滤波得到的。我们定义目标函数E{e²(n)}＝E{|x(n)-y(n)*f(n)|²}，这里*表示卷积，f(n)表示F(z)的时域脉冲响应。最小化该目标函数可得到时域的维纳解形式如下：As shown in FIG. 3 , FIG. 3 is a block diagram of the feedback path estimation system of the present invention. The signal picked up by the microphone is x(n)=d(n)+s(n), and d(n) is obtained by filtering the speaker output signal y(n) through the feedback path transfer function F(z). We define the objective function E{e ² (n)}=E{|x(n)-y(n)*f(n)| ² }, where * means convolution and f(n) means F(z) Time Domain Impulse Response. Minimizing the objective function can obtain the Wiener solution form in the time domain as follows:

$\overset{^^}{f f} ((n no)) = = {R R}_{yy yy}^{- - 11} {r r}_{yx yx} - - - - - - ((11))$

其中，R_yy表示y(n)的自相关矩阵，r_yx表示y(n)和x(n)的互相关矩阵。将(1)变换到频域得到Among them, R _yy represents the autocorrelation matrix of y(n), and _ryx represents the cross-correlation matrix of y(n) and x(n). Transform (1) into the frequency domain to get

${F f}^{' '} ((i i,, k k)) = = \frac{{P P}_{yx yx} ((i i,, k k))}{{P P}_{yy yy} ((i i,, k k))} - - - - - - ((22))$

这里P_yx(i，k)表示x(n)和y(n)的互功率谱，P_yy(i，k)表示y(n)的功率谱。如果输入信号s(n)和输出信号y(n)之间的延迟足够长，就可以很好的对语音这样的短时平稳信号进行解相关，即可以认为s(n)和y(n)之间的相关性已经变得很弱，那么P_ys(i，k)≈0，所以P_yx(i，k)＝P_ys(i，k)+P_yd(i，k)≈P_yd(i，k)，从而利用(2)式可以很有效的估计出反馈路径传递函数的频谱。本发明实际利用的是F′(i，k)的幅度谱|F′(i，k)|。如图5所示，图5为本发明的反馈频谱估计模块410的框图。Here P _yx (i, k) represents the cross power spectrum of x(n) and y(n), and P _yy (i, k) represents the power spectrum of y(n). If the delay between the input signal s(n) and the output signal y(n) is long enough, it can well decorrelate short-term stationary signals such as speech, that is, s(n) and y(n) can be considered The correlation between has become very weak, then P _ys (i, k)≈0, so P _yx (i, k)=P _ys (i, k)+P _yd (i, k)≈P _yd ( i, k), so that the spectrum of the feedback path transfer function can be estimated effectively by using formula (2). The present invention actually utilizes the magnitude spectrum |F'(i, k)| of F'(i, k). As shown in FIG. 5 , FIG. 5 is a block diagram of the feedback spectrum estimation module 410 of the present invention.

互功率谱平滑单元502计算x(n)和y(n)的互功率谱绝对值，并进行平滑，The cross power spectrum smoothing unit 502 calculates the absolute value of the cross power spectrum of x(n) and y(n), and performs smoothing,

φ_yx(i，k)＝α_Cφ_yx(i-1，k)+(1-α_C)|Y^*(i，k)X(i，k)| (3)φ _yx (i, k)=α _C φ _yx (i-1, k)+(1-α _C )|Y ^* (i, k)X(i, k)| (3)

这里使用*表示复共轭。我们采用了对信号周期图进行自回归平滑作为功率谱或互功率谱的估计，这主要是由于该方法简单和容易实现，然而本专利并不限于该方法，其他任何估计功率谱的方法都可以使用。Here * is used to indicate complex conjugation. We use the autoregressive smoothing of the signal periodogram as the estimation of the power spectrum or cross power spectrum, mainly because the method is simple and easy to implement, but this patent is not limited to this method, any other method for estimating the power spectrum can be use.

功率谱平滑单元504计算y(n)的功率谱并进行平滑，The power spectrum smoothing unit 504 calculates the power spectrum of y(n) and performs smoothing,

φ_yy(i，k)＝α_Rφ_yy(i-1，k)+(1-α_R)Y^*(i，k)Y(i，k) (4)φ _yy (i, k) = α _R φ _yy (i-1, k) + (1-α _R ) Y ^* (i, k) Y(i, k) (4)

这里α_C和α_R是平滑因子，其取值满足0≤α_C＜1及0≤α_R＜1。Here α _C and α _R are smoothing factors, and their values satisfy 0≤α _C <1 and 0≤α _R <1.

反馈传递函数估计单元506计算反馈路径传递函数幅度谱，表示为：The feedback transfer function estimation unit 506 calculates the magnitude spectrum of the feedback path transfer function, expressed as:

$| | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | = = \frac{{φ φ}_{yx yx} ((i i,, k k))}{{φ φ}_{yy yy} ((i i,, k k)) + + δ δ} - - - - - - ((55))$

式(5)中δ是一个绝对值很小的正数，用来防止式(5)的除数为0。In formula (5), δ is a positive number with a very small absolute value, which is used to prevent the divisor of formula (5) from being 0.

如图6所示，图6为一个实测的反馈路径脉冲响应图。如图7所示，图7为实际反馈路径传递函数的幅度谱和本发明估计出来的反馈路径传递函数的幅度谱对比图。可以看出本发明估计出的反馈路径传递函数的幅度谱能很好的逼近真实的反馈路径传递函数的幅度谱。As shown in FIG. 6, FIG. 6 is a measured feedback path impulse response diagram. As shown in FIG. 7 , FIG. 7 is a comparison chart of the magnitude spectrum of the actual feedback path transfer function and the magnitude spectrum of the feedback path transfer function estimated by the present invention. It can be seen that the amplitude spectrum of the feedback path transfer function estimated by the present invention can well approximate the real amplitude spectrum of the feedback path transfer function.

当前帧反馈信号幅度谱计算单元508用来估计当前帧反馈信号的频谱分量，表示为The current frame feedback signal amplitude spectrum calculation unit 508 is used to estimate the spectral component of the current frame feedback signal, expressed as

$| | \overset{^^}{X x} ((i i,, k k)) | | = = β β | | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | | | Y Y ((i i,, k k)) | | - - - - - - ((66))$

式(6)中，β是一个用户可调的参数，如果反馈信号的幅度谱被过估了则需要使β＜1，否则若反馈信号的幅度谱被欠估了则需要使β＞1。选择大的补偿因子就可以获得比较大的增益量但是相对的会对近端语音质量有一定影响；选择小的补偿因子会获得好的语音质量但是可能会减小扩声系统的增益量，应该根据实际情况进行合理的取舍。In formula (6), β is a user-adjustable parameter. If the amplitude spectrum of the feedback signal is overestimated, β<1 needs to be set; otherwise, if the amplitude spectrum of the feedback signal is underestimated, β>1 needs to be set. If you choose a large compensation factor, you can get a relatively large gain, but it will have a certain impact on the near-end voice quality; if you choose a small compensation factor, you can get good voice quality but may reduce the gain of the sound reinforcement system. Make reasonable trade-offs according to the actual situation.

综上所述，通过式(6)我们就估计出了反馈信号的幅度谱，完成了反馈信号频谱估计模块410的功能。To sum up, we can estimate the amplitude spectrum of the feedback signal through formula (6), and complete the function of the feedback signal spectrum estimation module 410 .

增益计算模块412的增益函数可以利用谱减法获得，其表达式是：The gain function of gain calculation module 412 can utilize spectrum subtraction to obtain, and its expression is:

$G G ((i i,, k k)) = = {[[\frac{max max (({| | X x ((i i,, k k)) | |}^{α α} - - {| | \overset{^^}{X x} ((i i,, k k)) | |}^{α α},, 00))}{{| | X x ((i i,, k k)) | |}^{α α}}]]}^{\frac{11}{α α}} - - - - - - ((77))$

上式中α表示谱减法的一个因子，α＝1是基于幅度谱减法，α＝2是功率谱减法即维纳滤波法，α也可以选择其他数值，α＞0。实际中为了减小计算复杂度一般选择α＝1或α＝2。In the above formula, α represents a factor of spectral subtraction, α=1 is based on amplitude spectral subtraction, α=2 is power spectral subtraction, that is, Wiener filtering method, α can also choose other values, and α>0. In practice, α=1 or α=2 is generally selected in order to reduce the computational complexity.

虽然增益函数的选择还有其它形式，如文献[1]中采用了基于先验信噪比的增益函数([1]Y.Ephraim and D.Malah，“Speech enhancement using a minimum mean squareerror short-time spectral amplitude estimator，”IEEE Trans.on Acoust.，Speech，SignalProcessing，vol.ASSP-32，pp.1109-1121，Dec.1984)，通过实验我们发现对于啸叫抑制系统使用后验的信号反馈比可以获得更好的结果，因而本发明推荐使用式(7)来计算增益函数。Although there are other forms of gain function selection, such as the gain function based on the prior signal-to-noise ratio ([1] Y.Ephraim and D.Malah, "Speech enhancement using a minimum mean square error short-time spectral amplitude estimator, "IEEE Trans.on Acoust., Speech, Signal Processing, vol.ASSP-32, pp.1109-1121, Dec.1984), through experiments we found that the signal-feedback ratio of the posterior using the howling suppression system can be A better result is obtained, so the present invention recommends using formula (7) to calculate the gain function.

步骤4)对麦克风信号频谱进行修正。Step 4) Correct the spectrum of the microphone signal.

频谱修正模块414的输入输出关系为：The input-output relationship of the spectrum correction module 414 is:

$\overset{^^}{S S} ((i i,, k k)) = = X x ((i i,, k k)) G G ((i i,, k k)) - - - - - - ((88))$

表示进行消除了反馈信号后的麦克风信号频谱。 Represents the spectrum of the microphone signal after the feedback signal has been eliminated.

步骤5)对麦克风信号频谱进行逆傅里叶变换得到输出。Step 5) Inverse Fourier transform is performed on the frequency spectrum of the microphone signal to obtain an output.

逆傅里叶变换模块416对当前帧进行逆傅里叶变换得到当前帧输出，然后与前一帧输出进行重叠叠加得到最终的输出信号 The inverse Fourier transform module 416 is for the current frame Perform inverse Fourier transform to obtain the current frame output, and then overlap with the previous frame output to obtain the final output signal

步骤6)可选的)对步骤5)的输出施加一个解相关模块418。Step 6) Optional) Apply a decorrelation module 418 to the output of step 5).

如步骤2)所述输入信号s(n)和输出信号y(n)之间的延迟是3倍的帧长，即对应50％的重叠。这个延迟一般可以对语音这样的短时平稳信号进行解相关。由于语音信号还具有长时相关性，即有时3倍帧长的延时仍然不足以完全去除其相关性，这可能会导致信号的失真。为了进一步提高语音质量需要增加一个解相关模块来去除s(n)和y(n)之间的相关性，很多经典的解相关方法已经被提出，可参见文献[2](M.Ali，“Stereophonic acoustic echo cancellation system using time-varying all-pass filteringfor signal decorrelation，”in Proceedings of the ICASSP’98，vol.6，pp.3689-3692，Seattle，USA，May 1998)和文献[3]([3]J.Benesty，D.R.Morgan，and M.M.Sondhi，“A betterunderstanding and an improved solution to the specific problems of stereophonic acousticecho cancellation，”IEEE Trans.Speech Audio Processing，vol.6，pp.156-165，Mar.1998)。需要说明的是，去相关算法的选取必须以不损伤语音质量为前提。The delay between the input signal s(n) and the output signal y(n) as described in step 2) is 3 times the frame length, ie corresponding to 50% overlap. This delay can generally decorrelate short-term stationary signals such as speech. Since the voice signal also has long-term correlation, that is, sometimes a delay of 3 times the frame length is not enough to completely remove its correlation, which may lead to signal distortion. In order to further improve the speech quality, it is necessary to add a decorrelation module to remove the correlation between s(n) and y(n). Many classic decorrelation methods have been proposed, which can be found in literature [2] (M.Ali, " Stereophonic acoustic echo cancellation system using time-varying all-pass filtering for signal decorrelation," in Proceedings of the ICASSP'98, vol.6, pp.3689-3692, Seattle, USA, May 1998) and literature [3] ([3 ] J. Benesty, D.R. Morgan, and M.M. Sondhi, "A better understanding and an improved solution to the specific problems of stereophonic acousticcho cancellation," IEEE Trans. Speech Audio Processing, vol. 6, pp. 156-165, Mar. 1998) . It should be noted that the selection of the decorrelation algorithm must be based on the premise of not damaging the voice quality.

步骤7)可选的)对步骤6)的输出增加一个增益模块420。Step 7) Optional) Add a gain module 420 to the output of step 6).

系统增益可以通过两种方式进行调节，一是调节数字信号的增益，即在增益模块420中对施加增益g，也可以通过调节模拟系统如功放的增益来实现。The system gain can be adjusted in two ways, one is to adjust the gain of the digital signal, that is, to adjust the gain in the gain module 420 Applying the gain g can also be realized by adjusting the gain of an analog system such as a power amplifier.

应该指出的是，本发明所描述的啸叫抑制方法可以用多种方式实现，例如硬件、软件或者是硬件和软件的组合。硬件平台可以是FPGA、PLD或其他专用集成电路ASIC。软件平台包括DSP、ARM或其他微处理器。软件和硬件的组合例如部分模块用DSP软件来实现，部分模块如FFT用硬件加速器来实现。It should be noted that the howling suppression method described in the present invention can be implemented in various ways, such as hardware, software or a combination of hardware and software. The hardware platform can be FPGA, PLD or other ASIC. Software platforms include DSP, ARM or other microprocessors. The combination of software and hardware, for example, some modules are realized by DSP software, and some modules such as FFT are realized by hardware accelerator.

最后所应说明的是，以上实施例仅用以说明本发明的技术方案而非限制。尽管参照实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，对本发明的技术方案进行修改或者等同替换，都不脱离本发明技术方案的精神和范围，其均应涵盖在本发明的权利要求范围当中。Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than limit them. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent replacements to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all of them should be included in the scope of the present invention. within the scope of the claims.

Claims

1. A howling suppression method based on feedback signal spectrum estimation, the specific steps of the method comprising:

Step 1): Framing, windowing and Fourier transform the signal x(n) collected by the microphone and the output signal y(n) of the speaker to obtain the frequency domain signal X(i,k) of the microphone and the frequency domain signal of the speaker output Y(i,k);

Step 2): According to the microphone frequency domain signal X (i, k) obtained in the step 1) and the loudspeaker output frequency domain signal Y (i, k), calculate the cross power spectrum φ _yx ( i,k) and the power spectrum φ _yy (i,k) of the loudspeaker output signal; and use the cross power spectrum φ _yx (i,k) and power spectrum φ _yy (i,k) to calculate the feedback path transfer function magnitude spectrum estimate Its calculation formula is expressed as:

| | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | = = \frac{{φ φ}_{yx yx} ((i i,, k k))}{{φ φ}_{yy yy} ((i i,, k k)) + + δ δ}

In the above formula, δ is a positive number with a very small absolute value;

Step 3): Using the feedback path transfer function amplitude spectrum obtained in step 2) to estimate Calculate the magnitude spectrum of the current frame feedback signal Its calculation formula is expressed as:

| | \overset{^^}{X x} ((i i,, k k)) | | = = β β | | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | | | Y Y ((i i,, k k)) | |

In the above formula, β is a user-adjustable parameter. If the amplitude spectrum of the feedback signal is overestimated, β<1 needs to be set; otherwise, if the amplitude spectrum of the feedback signal is underestimated, β>1 needs to be set;

Step 4): Using the microphone signal amplitude spectrum X(i,k) and the feedback signal amplitude spectrum obtained in step 3) Calculate the gain function G(i,k), and its calculation formula is expressed as:

G G ((i i,, k k)) = = {[[\frac{max max ((| | X x ((i i,, k k)) {| |}^{α α} - - | | \overset{^^}{X x} ((i i,, k k)) {| |}^{α α},, 00))}{{| | X x ((i i,, k k)) | |}^{α α}}]]}^{\frac{11}{α α}}

In the above formula, α represents a factor of spectrum subtraction, α=1 is based on amplitude spectrum subtraction, α=2 is power spectrum subtraction, that is, Wiener filtering method, α can also choose other values, α>0;

Step 5): According to the gain function G(i,k) obtained in the above step 4), the microphone frequency domain signal X(i,k) is corrected to realize howling suppression, and the formula for the correction is expressed as:

\overset{^^}{S S} ((i i,, k k)) = = X x ((i i,, k k)) G G ((i i,, k k))

In the above formula, Represents the spectrum of the microphone signal after the feedback signal has been eliminated.

2. The howling suppression method based on feedback signal spectrum estimation according to claim 1, characterized in that, the method further comprises:

Step 6): Process the corrected microphone frequency domain signal obtained in step 5) to obtain the final output signal and for the output signal performing a decorrelation process; and

Step 7): Perform gain adjustment on the processing result of step 6).

3. the howling suppression method based on feedback signal spectrum estimation according to claim 1, is characterized in that, described step 2) the specific step of obtaining feedback path transfer function amplitude spectrum comprises:

21): Using the first-order autoregressive smoothing method, the cross power spectrum φ _yx (i,k) is obtained according to formula (1);

φ _yx (i,k)=α _C φ _yx (i-1,k)+(1-α _C )|Y ^* (i,k)X(i,k)| (1)

Among them, α _C is a smoothing factor, 0≤α _C <1;

22): Using the first-order autoregressive smoothing method, the power spectrum φ _yy (i,k) is obtained according to formula (2);

φ _yy (i,k)=α _R φ _yy (i-1,k)+(1-α _R )Y ^* (i,k)Y(i,k) (2)

Among them, α _R is a smoothing factor, 0≤α _R <1;

23): Using the cross-power spectrum φ _yx (i,k) and the power spectrum φ _yy (i,k) to obtain the amplitude spectrum of the feedback path transfer function according to formula (3)

| {\hat{f}}^{'} (i, k) |;

| | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | = = \frac{{φ φ}_{yx yx} ((i i,, k k))}{{φ φ}_{yy yy} ((i i,, k k)) + + δ δ} - - - - - - ((33)) . .

4. the howling suppression method based on feedback signal spectrum estimation according to claim 1, is characterized in that, described step 3) calculates and obtains current frame feedback signal magnitude spectrum according to formula (4)

| | \overset{^^}{X x} ((i i,, k k)) | | = = β β | | {\overset{^^}{F f}}^{' '} ((i i,, k k)) | | | | X x ((i i,, k k)) | | - - - - - - ((44))

Among them, β is an adjustable parameter, which should be reasonably selected according to the actual situation.

5. The howling suppression method based on feedback signal spectrum estimation according to claim 1, characterized in that, the method for calculating the gain function in the step 4) comprises: based on a priori signal feedback ratio method and based on a posteriori signal feedback than method.

6. The howling suppression method based on feedback signal spectrum estimation according to claim 5, characterized in that, described step 4) calculates and obtains the gain function G(i, k);

G G ((i i,, k k)) = = {[[\frac{max max ((| | X x ((i i,, k k)) {| |}^{α α} - - | | \overset{^^}{X x} ((i i,, k k)) {| |}^{α α},, 00))}{{| | X x ((i i,, k k)) | |}^{α α}}]]}^{\frac{11}{α α}} - - - - - - ((55))

Wherein, α>0.

7. The howling suppression method based on feedback signal spectrum estimation according to claim 1, characterized in that, said step 5) is implemented by modifying the microphone frequency domain signal X(i,k) according to formula (6) Howling suppression:

\overset{^^}{S S} ((i i,, k k)) = = X x ((i i,, k k)) G G ((i i,, k k)) - - - - - - ((66)) . .

8. The howling suppression method based on feedback signal spectrum estimation according to claim 2, characterized in that the step 7) adopts adjusting the gain of the digital signal or adjusting the gain of the power amplifier of the analog system.