KR101306868B1 - Un-identified system modeling method and audio system for howling cancelation using it - Google Patents

Abstract

The present invention estimates the echo channel using the frequency domain, but executes the estimated filter in the time domain to reduce the amount of calculation and to quickly generate the echo component to remove it without time delay, and to accurately adapt the echo component to changes in the surrounding environment. The present invention relates to an unidentified system modeling method that improves the sound collection performance by estimating and leveling the input signal to a microphone, and to a howling cancellation audio system using the same. And with echo cancellation performance, you can also realistically listen to his voice accurately without noise in the field of immersive motion games.

Description

UN-IDENTIFIED SYSTEM MODELING METHOD AND AUDIO SYSTEM FOR HOWLING CANCELATION USING IT}

The present invention estimates the echo channel using the frequency domain, but the estimated filter is executed in the time domain to reduce the amount of calculation and to quickly generate the echo component to remove without time delay, and to estimate the echo channel by adapting to changes in the surrounding environment. In addition, the present invention relates to an unidentified system modeling method for improving sound collection performance by adjusting a level of a signal input to a microphone, and a howling cancellation audio system using the same.

An audio system that amplifies the sound input to the microphone and outputs it to the speaker is generally used to convey the sound coming from the speaker or the microphone's sound collection area to a large number of people or the speaker himself.

However, in such an audio system, howling occurs due to positive feedback amplification by an echo phenomenon in which sound output from a speaker is input into a microphone. In order to remove such howling, echo may be suppressed only by estimating an echo channel leading to a microphone from a speaker and removing an echo component from a signal input to the microphone.

As a technique of suppressing echo components by estimating echo channels, a technique of removing echo components by processing a signal in a frequency domain has been disclosed.

8 is a block diagram of an echo cancellation system disclosed in Korean Patent No. 10-0566452. The above-described prior art is a technology applied in a communication system between far-end and near-end, but is also applicable to an audio system. According to this, as the echo channel is modeled using a filter in the frequency domain, the signal x to be output through the speaker is Fourier transformed to calculate the echo component in the frequency domain, and then the inverse Fourier transform is used to obtain the echo component in the time domain. A method of removing one echo component from a signal y input to a microphone is disclosed. However, the above-described prior art is configured to perform the Fourier transform (FFT) three times and the Inverse Fourier transform (IFFT) two times in order to remove the echo component, so that a large amount of calculation is required, and furthermore, to update the filter in the frequency domain. If you look at the calculation process to obtain the echo component except for the process, it is necessary to perform Fourier transform (FFT, 210) and filter (215) and then inverse Fourier transform (IFFT, 225) to output the signal through the speaker. The calculation amount is larger than that of the time domain filter, and accordingly, a time delay occurs.

In order to solve such a problem due to the amount of computation and the time delay, that is, the problem of failing to remove the echo component in time from the micro-input signal, it is difficult to configure the high-speed processing.

On the other hand, when looking at the types of microphones, microphones used close to the speaker's mouth, such as dynamic microphones and Goose-Neck microphones, and sound collection characteristics, such as condenser microphones and Bluetooth microphones, are close to the speaker's mouth. There is no need to distinguish between the microphones. Here, the microphone used in close proximity is strong in echo noise, but inconvenient to use, and the microphone having excellent sound collection characteristics can also collect sound from a long distance, so that the speaker can use without inconvenience, but is vulnerable to echo noise. Accordingly, in order to reduce the reverberation, the speaker had to wear a microphone while wearing the discomfort or wear a fixing means for fixing the microphone near the mouth.

For example, if you look at the field of tactile motion games, you can enjoy the game only if you listen to your voice again, but because of the nature of the motion game, you have to move your body, so holding a microphone or wearing a fixed means restricts the movement in the motion game. It also interferes with playing the game.

Therefore, there is a limit to improving both the sound collection performance and the echo suppression performance at the same time by the microphone itself. Therefore, an audio system that has excellent sound collection performance and strong echo noise is used by using a microphone that does not need to be held by hand or worn on the head. Required.

In addition, if you look at the actual usage of the audio system, if the surrounding situation changes, such as moving the position of the microphone or speaker, placing a new object around, moving an object away, hanging a curtain, walking, etc. As a result, the echo channel from the speaker to the microphone changes little by little. Thus, there is a need for an audio system that can adapt to and remove echoes even when the surrounding environment changes.

KR 10-0566452 B1 2006.03.31. KR 10-2009-0110069 A 2009.10.21.

Accordingly, an object of the present invention is to determine an unidentified channel by processing a signal in the frequency domain, while reducing the amount of computation compared to the prior art, while unfiltered system modeling method for quickly filtering with a small amount of computation when filtering with the estimated channel and It is to provide a howling elimination audio system.

Another object of the present invention is to provide a howling cancellation audio system that satisfies the sound collection performance and the echo cancellation performance simultaneously.

It is still another object of the present invention to provide a howling cancellation audio system that actively adapts to changes in the surrounding environment to eliminate echoes.

In the present invention for achieving the above object, in the audio system for outputting the sound signal of the microphone to the speaker, after filtering the output signal input to the speaker with a filter 21, which estimates the echo channel between the speaker and the microphone, Howling removal means (20) for outputting an error signal obtained by subtracting the filtered signal from the acoustic signal of the microphone from the speaker; Fourier transform signal and Fourier transform signal and Fourier transform (conjugate) signal of the speaker output signal are multiplied by the sample unit according to the sample order of the signal coefficient in the frequency domain After obtaining, the inverse Fourier transform is performed to obtain the filter coefficient update component of the time domain, and a value obtained by multiplying a predefined update gain by the filter coefficient update component of the time domain is added to or subtracted from the coefficient of the filter. Filter coefficient estimation means (30) for updating the coefficient of the filter; Characterized in that configured to include.

In the filter coefficient estimating means 30, a Fourier transform is a Fast Fourier Transform (FFT), and an Invers Fourier Transform is an Inverse Fast Fourier Transform (IFFT). It is characterized by that.

The howling removing means 20 includes a delay unit 22 for delaying the output signal of the speaker and inputting it to the filter by the time required for the output signal of the speaker to be echoed to the microphone. The filter coefficient estimating means 30 ) Is characterized by using the signal delayed by the delay unit 22 as a speaker output signal.

The filter coefficient estimating means 30 has a spectrum analyzer 35 for analyzing the frequency component of the error signal, and updates the coefficient of the filter when power for each frequency is dispersed to a predetermined uniformity. do.

It is characterized in that the coefficients of the filter are updated for a predetermined time initially at power-on of the howling cancellation audio system.

The spectrum analyzer 35 may calculate a dispersion degree of power within a frequency occupying a predetermined upper power range and update the coefficient of the filter when the uniformity is satisfied.

The uniformity is defined as a ratio value, and the spectrum analyzer 35 has a power average at a frequency occupying a predefined first upper power range and a next-order power range with respect to the predefined first upper power range. After calculating the power average at the frequency occupying the second upper power range defined by the, characterized in that the coefficient of the filter is updated when the calculated both power average is less than the uniformity.

Further comprising a power adjusting means 40 for adjusting the power of the sound signal received through the speaker, the power adjusting means 40, the mapping characteristics of the input signal and the output signal to form a gamma curve to provide a low power signal And a power level adjustment filter for lowering the signal of a high power, and a peak window filter for alleviating and outputting peaks in the input signal.

In addition, the present invention for achieving the above object, in the modeling method for estimating the coefficient of the filter to be connected to the input terminal and the output terminal of the unidentified system, the signal obtained by filtering the input terminal signal of the unidentified system with the filter An error signal obtaining step of obtaining an error signal by subtracting from the output terminal signal of the unidentified system; A frequency domain error signal obtaining step of Fourier transforming the error signal; A frequency domain input signal obtaining step of Fourier transforming an input signal input to an unidentified system through an input terminal; A frequency domain coefficient update component obtaining step of obtaining a frequency domain filter coefficient update component based on the frequency domain input signal and the frequency domain error signal; A time domain coefficient update component obtaining step of obtaining a time domain filter coefficient update component by inverse Fourier transforming the frequency domain filter coefficient update component; A filter coefficient updating step of adding or subtracting a value obtained by multiplying a predefined update gain by a time domain filter coefficient update component from coefficients of the filter; An update persistence determining step of returning to the error signal obtaining step and repeatedly updating a filter coefficient until the preset condition is satisfied; And a control unit.

The frequency domain coefficient update component acquiring step may include obtaining a frequency domain filter coefficient update component by multiplying a signal obtained by a conjugate operation on a frequency domain input signal and a frequency domain error signal in a sample unit according to the sample order of the signal.

The input signal in the error signal acquiring step and the coefficient update component acquiring step may be a signal obtained by delaying an input terminal input signal by a delay time of an unidentified system.

The present invention as described above uses a filter in the time domain, but obtains an update component of the filter coefficient in the frequency domain, thereby reducing the number of Fourier transforms and inverse Fourier transforms, thereby reducing the amount of computation compared to the method of eliminating echo in the conventional frequency domain. In addition, since there is no Fourier transform and an Inverse Fourier transform in the filtering process of the signal, fast filtering is possible.

In addition, the present invention improves the sound collection performance and echo cancellation performance of the microphone by adjusting the level of the sound signal with the power adjusting means 40, and thus, the pulpit audio system, conference room using a sound collecting microphone that is not worn like a condenser microphone It can be applied to an audio system without any performance deterioration. Furthermore, even if it is applied to an audio system in the field of haptic motion games that a person moves and enjoys, he can listen to his voice without noise and enjoy it realistically.

In addition, the present invention adapts and estimates the fluctuation of the echo channel according to the change of the surrounding environment, thereby accurately removing the echo component, and further, selectively estimates the echo channel using the spectrum analyzer 35 of the filter coefficient estimating means. Therefore, there is no fear of estimation error and no fear of divergence.

1 is a block diagram of a howling cancellation audio system according to an embodiment of the present invention.
2 is a spectral graph of an acoustic signal illustrated to illustrate environmental adaptation in an embodiment of the present invention.
3 is a graph of a gamma curve attached to illustrate power adjustment of an acoustic signal in an embodiment of the present invention.
4 is a signal graph attached to illustrate peak suppression in an embodiment of the present invention.
5 is a flow chart of a howling removal method made by a howling cancellation audio system according to an embodiment of the invention.
FIG. 6 is a detailed flowchart of an initialization adaptation step in the flowchart of FIG. 5, illustrating a method of modeling an unidentified system according to an embodiment of the present disclosure. FIG.
7 is a detailed flowchart of the environmental adaptation step during normal operation in the flowchart of FIG.
8 is a block diagram of an echo canceller according to the prior art.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

1 is a block diagram of a howling cancellation audio system according to an embodiment of the present invention.

The howling cancellation audio system according to the embodiment of the present invention shown in FIG. 1 includes: sound processing means (10) for outputting a sound signal received through the microphone (11) through the speaker (12); Howling removal means 20 for removing the echo component from the sound signal; And a filter coefficient estimating means 30 for estimating an echo channel for removing an echo signal.

Since the sound processing means 10 is a configuration employed in a general audio system, briefly explaining only basic components, the microphone 11 converts sound into an electric signal, and the analog sound signal converted from the microphone 11 is digital sound. A digitizer 13 for converting a signal into a signal, a signal processor 15 for amplifying a digital sound signal or other stereoscopic signals, and an analogizer for converting a digital sound signal transmitted from the signal processor 15 into an analog signal. 14, and a speaker 12 for outputting an analog signal as sound. Since the sound processing means 10 may be variously configured according to the type of the audio system, a detailed description thereof will be omitted.

Howling is generated when the sound is input through the microphone by the sound processing means 10 configured as described above and expanded through the speaker. In the howling, the sound output from the speaker through the echo channel 16 between the speaker and the microphone is resonated or oscillated by a repetitive positive feedback loop in which the sound output from the speaker is input to the microphone as an echo signal and amplified through the speaker. Say that.

Referring to FIG. 1, a signal output from a speaker is reflected through an echo channel 16 and input through a microphone along with a speaker's voice. The signal obtained by filtering the signal output from the speaker with the filter 21 is subtracted from the signal input to the microphone. Accordingly, the sound signal input through the microphone is removed through the echo channel 16 and output through the speaker, thereby suppressing howling. In addition, in the present invention, when the echo channel 16 is modeled by the filter 21 in the time domain, the coefficient (or tap) of the filter 21 is updated with an update component obtained in the frequency domain. The howling removing means 20 and the filter coefficient estimating means 30 for this purpose will be described.

The howling removal means 20 is a delay unit 22 for delaying an output signal (precisely a digital signal input to the analogization unit, u ') of the speaker 12, and a delay unit as a model for estimating the echo channel 16. A filter 21 for filtering the signal u delayed by the 22 and a signal filtered with the filter 21 are subtracted from an acoustic signal of the microphone 11 (exactly, the signal output from the digitizer, d). It is comprised including the part 23.

In this case, the delay unit 22 may delay the signal delay time of the path to the track environment from the input terminal of the analogization unit 14 to the subtraction unit 23 via the echo channel 16. This is because the delay time of the signal is generated not only in 16) but also by the digitizing unit 13 and the analogizing unit 14. In addition, the delay unit 22 may reduce the number of taps of the filter 21, and further, filter coefficient estimating means (C) as a correlation between the original signal (signal output to the speaker, u ') and the error signal (e). 30) to make an accurate estimate.

Since the filter 21 uses digital data as signal processing as a filter in the time domain, the filter 21 becomes a filter having a finite length of taps, and the filter coefficients are updated by the filter coefficient estimating means 30 to be described later. Will be modeled.

The subtractor 23 generates an error signal (or residual signal, e) obtained by subtracting the output signal y of the filter 21 from the sound signal d of the microphone so that the sound signal d of the microphone is generated. Instead, the error signal e is output from the speaker 12 as sound. In other words, the subtractor 23 removes the echo component y estimated by the filter 21 even when the sound of the external environment is input together with the sound of the echo.

The filter coefficient estimating means 30 is a means for updating the coefficient of the filter 21 in the time domain to adapt to the echo channel 16. The filter coefficient estimating means 30 discretely modulates the error signal e that is an output signal of the subtractor 23. A first Fourier transform unit 34 performing a Discrete Fourier Transform; A spectrum analyzer 35 for analyzing the spectrum of the error signal in the frequency domain output from the first Fourier transformer 34 and determining whether to update the filter coefficients; A second Fourier transform unit 36 for Discrete Fourier Transform of the speaker 2 output signal u output from the delay unit 22; A power normalization unit 37 for normalizing the speaker output signal in the frequency domain output from the second Fourier transform unit 36; A multiplier 33 for extracting filter coefficient update components in the frequency domain based on the error signal in the frequency domain and the speaker output signal in the normalized frequency domain; An inverse Fourier transform unit 32 for performing discrete inverse fourier transform on the filter coefficient update component in the frequency domain by the multiplier 33 to obtain a filter coefficient update component in the time domain; And a filter coefficient updater 31 for updating the coefficients of the filter 21 by reflecting the filter coefficient update component in the time domain. .

The first Fourier transform unit 34 and the second Fourier transform unit 36 convert a signal in the time domain into a signal in the frequency domain in discrete signal processing, and reduce the number of operations. : Fast Fourier Transform). In addition, the inverse Fourier transform unit 32 converts a signal in the frequency domain into a signal in the time domain in discrete signal processing, and preferably includes an inverse fast Fourier transform (IFFT).

Here, the fast Fourier transform converts 2 ^k signal samples whose k is an integer, and the inverse fast Fourier transform converts signal samples having the same length as the fast Fourier transform. The larger the number of taps (the length of the filter or the number of coefficients) of the filter 21 is, the longer the echo channel 16 can be modeled. However, the length of the tap 21 is appropriate in consideration of the use environment and implementation complexity. In general, the number of taps in a filter is half the length of the number of taps in the fast Fourier transform.

In a specific embodiment of the present invention, the first Fourier transform unit 34 takes an error signal e as many times as the number of taps of the filter 21 and inputs "0" by a difference from the number of taps of the Fourier transform. Fourier transform after The second Fourier transform unit 36 takes the speaker output signal by the number of taps of the Fourier transform and performs Fourier transform as it is. Next, since the time domain signal converted by the inverse Fourier transform unit 32 is generated by the number of taps of the inverse Fourier transform, only the number of taps of the filter 21 is taken from the converted time domain signal, and the filter coefficient update component of the time domain is taken. Get The position of the sample taken here is selected based on the zero insertion by the first Fourier transform section 34.

The power normalization unit 37 is a known technique for normalizing the power of a signal corresponding to the number of taps of a Fourier transform. For example, the power normalization unit 37 reflects newly input signal power in the process of repeatedly updating the filter coefficients. It is to adjust the ratio, which can be implemented as a known technique, and thus detailed description is omitted.

The multiplier 33 multiplies the sample of the error signal E in the frequency domain with the samples obtained by conjugating the speaker output signal samples in the frequency domain, one-to-one, according to the sample order, and multiplies the sample unit. This generates a signal having a length corresponding to the number of taps of a Fourier transform.

The resultant signal by multiplication in the multiplier 33 is a frequency domain signal of the filter coefficient update component, which will be described below.

In modeling an unknown system as a filter in a time domain, an input signal input to an unknown system is u, an output signal output from an unknown system d, and a signal output from the filter when the input signal u is input to the filter. If y is the error signal obtained by subtracting the output signal y of the filter from the output signal d of the unknown system, e and the vector of the filter coefficients is W, the block LMS algorithm (Block Least-Mean-Square Algorithm) The update expression of the used filter coefficients is expressed as follows.

는 갱신 이득이고, n은 알고리즘의 반복 스텝이고, W_-1 는 이전의 반복 스텝에서 얻은 필터 계수의 벡터이고, W는 현재의 반복 스텝에서 얻게 되는 필터 계수의 벡터이며, 상기 업데이트식에서 대문자(capital)로 표기한 것은 벡터(vector)이다.Where L is the block size in the block LMS algorithm,

Is the update gain, n is the iteration step of the algorithm, W _-1 is the vector of filter coefficients obtained from the previous iteration step, W is the vector of filter coefficients obtained from the current iteration step, Indicated as) is a vector.

Looking at the above update expression, which is the update component of the filter coefficient in the time domain

Is the convolution operation of the error signal e and the input signal u, and the convolution in the time domain is multiplied by the error signal in the frequency domain and the input signal in the frequency domain by the characteristic corresponding to the multiplication in the frequency domain. You can get it. The convolution is calculated by inverting any one of the signals on the time-axis when calculating in the time domain. However, since the frequency components are equally obtained even if the signals in the time domain are inverted, the update component of the filter coefficients is calculated. It can be seen that can be obtained by multiplication in the frequency domain.

In the present invention, the multiplier 33 multiplies the frequency signal of the error signal e by the frequency signal of the speaker output signal (precise normalized and then conjugated) by a sample unit to update the filter coefficients in the frequency domain. Obtain the component. In this case, "multiplying by sample unit" means multiplying elements in the same row in both column vectors to obtain column vectors having the same length as both column vectors, and converting the column vector of the error signal (e) in the frequency domain to E. When the column vector of the speaker output signal is U, it is obtained as a diagonal component in the matrix of EU ^H. Here, the subscript H is an operation including a transpose and a conjugate as a Hermitian operator.

Next, the inverse Fourier transform unit 32 converts a column belt extracted only a diagonal component from the matrix of EU ^H into an inverse Fourier transform and converts the signal into a time domain signal. The filter coefficient update component of the time domain is obtained by extracting only sample elements of the correct length. Here, a conjugate may also be executed in correlation with the filter coefficients.

The filter coefficient updating unit 31 updates the filter coefficients by reflecting the filter coefficient updating component of the time domain obtained by the inverse Fourier transform unit 32 to the filter 21. At this time, the update of the filter coefficient is a value obtained by multiplying the predefined update gain by the filter coefficient update component, and the update gain is related to the convergence speed when modeling the filter. Select.

The spectrum analyzer 35 analyzes the frequency components of the error signal e so as to update the coefficients of the filter when the power for each frequency is dispersed above a predetermined uniformity. It is possible to accurately estimate the echo channel 16 by updating the filter coefficients with a random training signal with a uniform frequency distribution. However, when the filter coefficients are updated in accordance with the environment as described in detail below, This is because the echo channel 16 cannot be accurately estimated if the sound of the surrounding environment input through the microphone 11 is a signal biased at a specific frequency.

Referring to the power spectrum of the different acoustic signals illustrated in FIG. 2, a in FIG. 2 is an acoustic signal in which power is large at a specific frequency, and b in FIG. 2 is an acoustic signal having a generally uniform value of power for each frequency across frequencies. . Here, as shown in b of FIG. 2, when the filter coefficient is updated when an acoustic signal having a uniform frequency component is input, the echo channel 16 can be estimated accurately, but as shown in a of FIG. If the filter coefficient is updated when an acoustic signal showing a large power is input, the reflection coefficient 16 may be updated due to a biased frequency component having a large power, and thus the echo channel 16 may not be accurately estimated. Will diverge.

According to an exemplary embodiment of the present invention, the spectrum analyzer 35 calculates a degree of power dispersion at a frequency corresponding to a predetermined upper power range, compares the uniformity, and then compares the uniformity with the filter only when the uniformity is satisfied. The coefficient updater 31 causes the coefficients of the filter to be updated.

In addition, according to a specific embodiment of the present invention, in order to easily calculate the degree of dispersion of the power, the uniformity is set as a ratio value, and the spectrum analyzer 35 at a frequency occupying a first predetermined upper power range. After calculating a power average at and a frequency occupying a second upper power range defined as a next-order power range with respect to the predefined first upper power range, the degree of dispersion is calculated by the ratio of the calculated power averages. The filter coefficient updating unit 31 updates the coefficient of the filter when both power averages differ by less than the uniformity, by comparing the obtained dispersion degree with the uniformity.

On the other hand, the sound processing means 10 is to remove the echo component estimated by the howling removal means 20 from the sound signal input to the microphone 11 to adjust the power of the sound signal to transmit to the speaker (2) A power adjusting means 40 is further provided.

The power adjusting means 40 includes a power level adjusting filter 41 for suppressing small sounds and a peak window filter 42 for suppressing peaks.

의 조건하에

의 특성으로 입력신호의 파워를 변형하여 출력하여서, 작은 소리를 더욱 낮게 하고 큰 소리를 더욱 크게 한다. 이에 따라, 마이크가 향하는 방향의 집음 공간에서 나는 소리를 크게 하는 대신에 집음 공간에서 벗어난 주변의 잡음을 더욱 낮게 하여서 집음(集音) 성능을 높일 수 있다. 또한, 상기 필터계수 추정수단(30)으로 반향채널(16)을 추정한다 해도 반향채널(16)과 동일할 정도로 추정할 수는 없어서 하울링 제거수단(20)으로 제거하지 못한 미약한 반향성분이 스피커(12)를 통해 출력될 수 있으나, 상기 파워레벨 조정필터(41)에 의해서 미약한 반향성분을 더욱 억제하므로 하울링 억제 성능을 높인다.Specifically, the power level adjusting filter 41 has a gamma curve between the input signal and the output signal to make the low power signal lower and the high power signal higher. Referring to the gamma curve illustrated in FIG. 3,

Under the terms of

By transforming the output power of the input signal with the characteristic of lowering the small sound and making the loud sound more louder. Accordingly, instead of making the sound produced in the sound collecting space in the direction toward which the microphone is directed, the sound collection performance can be improved by lowering the ambient noise outside the sound collecting space. In addition, even if the echo channel 16 is estimated by the filter coefficient estimating means 30, the weak echo component cannot be estimated to be the same as the echo channel 16 and thus cannot be removed by the howling removal means 20. Although it may be outputted through 12, the weak leveling component is further suppressed by the power level adjusting filter 41, thereby improving howling suppression performance.

The peak window filter 42 is a filter that mitigates and outputs a peak in an acoustic signal (exactly, an error signal passing through the power level adjustment filter). Referring to FIG. 4, when a peak occurs in a source signal, the signal is output as a signal below a predetermined level by covering a window centered on a sample having a peak. Here, the window has a lower value gradually toward the front and rear samples with the peak as a peak, and the sample having the peak is set to a preset level, and the samples before and after the peak are gradually smaller than the level adjustment amount of the peak.

Thus, the peak window filter 42 for mitigating the peak may be input when the speaker is very close to the microphone, for example, when the speaker is very close to the microphone. Reduce the sound and output to the speaker.

The power adjusting means 40 composed of the above-described power level adjusting filter 41 and the peak window filter 42 has an installation position between the subtracting part 23 and the signal processor 15 as shown in FIG. It is not limited, but may be provided at the rear end of the signal processor 15 or the front end of the subtraction section 23.

Meanwhile, the audio system illustrated in FIG. 1 includes a random signal input unit 50 for inputting a training signal r. The random signal input unit 50 is used to update the coefficient of the filter 21 initially after the audio system is installed or when the audio system is turned on, and according to an embodiment of the present invention, It is input to the front end and synthesized into an acoustic signal input to the microphone 1. The training signal r input here preferably comprises a pseudo-random signal in which all frequency components are evenly distributed.

In the exemplary embodiment of the present invention, the random signal input unit 50 is a component included in the sound processing means 10. However, the echo channel 16 may be modeled by updating the coefficients of the filter 21 at the initial installation of the audio system. When the echo channel 16 is updated using only a sound signal and then using an acoustic signal input to the microphone, the random signal input unit 50 may be separated. In addition, the random signal input unit 50 may be connected to the front end of the signal processor 15 or the rear end of the signal processor 15 instead of the front end of the subtractor 23. To correspond to the sound signal input to the microphone, it is preferable to connect to the front end of the subtractor 23 as shown in FIG.

The howling cancellation audio system according to the embodiment of the present invention has been described above, and a howling removal method including the howling removal audio system will be described. In addition, the unconfirmed system modeling method according to the embodiment of the present invention is described in the initialization adaptation step (S2), the initial environment adaptation step (S4) or the normal operation environment adaptation step (S5) described with reference to FIGS. Since it can be described as a matter, it is omitted to describe the accompanying drawings. In addition, since the howling removal audio system has been described in detail above, the following howling removal method will be briefly described.

5 is a flowchart of a howling removal method according to an embodiment of the present invention, FIG. 6 is a detailed flowchart of the initialization adaptation step S2 in the flowchart of FIG. 5, and FIG. 7 is an environment adaptation step during normal operation in the flowchart of FIG. 5. It is a detailed flowchart of (S5).

First, referring to FIG. 5, in the howling removal method, after installing the above-described audio system (S1), the echo channel 16 is estimated by the filter 21 by an initialization adaptation step S2. When the audio system is executed to use the system (S3), the initial environment adaptation step (S4) is performed at the beginning of the execution to reflect the change in the echo channel 16 according to the change of the surrounding environment in which the audio system is installed in the filter 21. After that, run the audio system normally. In addition, if the quality of the sound signal input to the microphone satisfies the conditions set forth in the present invention even during normal operation of the audio system, the filter 21 is updated using the sound signal, thereby continuously adapting to changes in the surrounding environment.

Hereinafter, the initialization adaptation step S2, the initial environment adaptation step S4 and the environmental adaptation step S5 during normal operation will be described with reference to the drawings.

In the initialization adaptation step (S2), as shown in FIG. 6, after the setting of the echo channel delay time (S10), the training signal r is input (S11) and the filter 21 is input to the echo channel (S11). 16 to model the echo channel 16 as a filter 21.

Here, in the setting of the echo channel delay time (S10), the delay time of the echo path leading to the speaker 12, the echo channel 16, and the microphone 11 is transmitted to the delay unit 22 of the howling removal means 20. This step is to set up. The delay time set here is, for example, prior to inputting the training signal of the random signal input unit 50 for the estimation of the echo channel 16, the training signal r of the random signal input unit 50 is separately performed. Input and detect the time difference between the output signal of the speaker 12 and the input sound signal of the microphone 11 can be set.

The training signal is input (S11) by inputting a training signal using the random signal input unit 50 and outputting the sound through the speaker 12, and the sound is input to the microphone 11 via the echo channel 16. In this case, the reflection channel 16 having the output signal of the speaker 12 as an input and the input signal of the microphone 11 as an output is modeled by the filter 21.

The modeling of the filter 21 is based on a terminal into which an output signal is input from the speaker 12 as an input terminal, and a reverberation channel using a terminal of a microphone 11 to which an acoustic signal of a sound sensed by the microphone 11 is output. Modeling the unidentified system including the filter 16 by the filter 21, the error signal acquisition step (S20), the frequency domain error signal acquisition step (S21), the frequency domain input signal acquisition step (S30, S31), frequency domain A coefficient update component acquisition step (S40), a time domain coefficient update component acquisition step (S50), a filter coefficient update step (S60), and an update duration determination step (S70) are included.

The error signal acquiring step (S20) is a step of obtaining an error signal by subtracting a signal filtered by the input terminal signal with the filter 21 from the output terminal signal, in the embodiment of the present invention by the delay unit 22 the input terminal signal Is delayed by the delay time of the unidentified system and then filtered by the filter 21. Here, the error signal obtaining step is to obtain an error between the output of the unidentified system and the output of the filter for the same input.

The frequency domain error signal acquisition step S21 is a step of converting the error signal into a signal in the frequency domain by Fourier transforming the error signal by the first Fourier transform unit 34. Here, since there is a difference between the number of samples of the error signal and the number of taps of the first Fourier transform unit 34, zero insertion is performed by the difference.

The frequency domain input signal acquiring step (S30, S31) is a step of obtaining a frequency domain signal of the input terminal signal by Fourier transforming the input terminal signal to the second Fourier transform unit 36. In the embodiment of the present invention, the input terminal signal delayed by the delay unit 22 is Fourier transformed, and the signal in the frequency domain obtained by the Fourier transform is normalized by the power normalization unit 37.

The frequency domain coefficient update component acquiring step (S40) comprises performing an error signal (operation including pre- and conjugation operations) after inputting a signal from an error signal and an input signal obtained from a signal of a frequency domain by Fourier transform. Multiplying the input stage signal obtained by the helical operation, and obtaining the coefficient update component obtained by multiplying the error signal and the input terminal signal (a signal obtained by the conjugate operation of the input terminal signal) in units of samples. The coefficient update component obtained here is a component in the frequency domain.

The time domain coefficient update component obtaining step (S50) is a step of inverse Fourier transforming the coefficient update component of the frequency domain obtained above, thereby obtaining the coefficient update component of the time domain. Here, there is a difference in the number of taps of the inverse Fourier transform and the Fourier transform and the number of taps of the filter, and among the components of the inverse Fourier transform, only the component corresponding to the number of taps of the filter is taken to obtain a coefficient update component.

The filter coefficient updating step (S60) is a step of adding or subtracting a value obtained by multiplying a predefined update gain by the coefficient update component of the time domain to the coefficient of the filter 21.

The determination of whether to continue the update (S70) is a step of returning to the error signal obtaining step (S20) and repeatedly updating the filter coefficients until the preset condition is satisfied. The predetermined condition may be, for example, when the input of the training signal is terminated by the random signal input unit 50 when the power of the error signal becomes less than or equal to a preset value.

As described above, by modeling the unidentified system including the echo channel to the filter 21 by the initialization adaptation step (S2), the echo component is removed from the sound signal input to the microphone when using the audio system and then output to the speaker. Thus, howling can be suppressed.

However, even after the audio system is installed and initialized and adapted as described above, the echo channel may change little by little due to changes in the surrounding environment. For example, the reflection channel may change little by little due to a change in the position of a microphone or a speaker, a movement of an object such as a table or a desk in a space where an audio system is installed, the appearance of a new object, a change by a curtain or a walk.

Accordingly, in the present invention, when the audio system adapted to the echo channel is turned on in the initialization adaptation step (S2), the initial environmental adaptation step (S4) is performed, and after the normal operation of the audio system, the environmental adaptation step ( S5) is performed.

The initial environment adaptation step (S4) is a step of adapting the echo channel 16 to the training signal r by operating the random signal input unit 50 for a few seconds at the beginning of turning on the audio system. Compared to the initialization adaptation step S described above with reference to FIG. 6, only the difference does not include the echo channel delay time setting step S10 and the echo channel delay time setting step S10 in the initialization adaptation step 2S. This is the same order that will be executed later. In other words, the initial environmental adaptation step S4 is executed for a preset time without setting a delay time at the initial time of turning on the audio system, and the echo channel 16 that varies according to the environmental change after the initial adaptation step S2 is performed. Is adapted to the filter 21.

The environmental adaptation step (S5) during the normal operation is a predetermined sound signal input to the microphone during the normal operation (that is, the operation of amplifying the microphone input sound and output to the speaker) after the initial environment adaptation step (S4) A step of removing echoes while adapting the filter 21 to the echo channel 16 when the quality is satisfied, as shown in FIG. 7, setting a delay channel delay time (S10) and training signal input step (S11). It differs from the initialization adaptation step S2 in that it does not perform. This is because the echo channel delay time does not fluctuate greatly and the filter coefficients are updated by the sound of the microphone sound, rather than using a separate training signal.

In addition, the environmental adaptation step (S5) during the normal operation is different from the initialization adaptation step (S2) of FIG. 6 in that it is selectively updated according to the quality of the sound signal when the filter coefficients are updated, and the filter coefficients are updated. The process may include the error signal acquisition step S20, the frequency domain error signal acquisition step S21, the frequency domain input signal acquisition steps S30 and S31, and the frequency domain coefficient update component acquisition step in the initialization adaptation step S2. S40), time domain coefficient update component acquisition step S50, and filter coefficient update step S60.

Referring to such a difference, while the echo signal is removed from the micro-input acoustic signal by the filter 21 adapted to the initial environmental adaptation step (S4) and then amplified and output to the speaker, an error signal ( Acquiring the echo component of the microphone from the echo signal) and the output signal of the speaker successively to perform the frequency domain error signal acquisition step (S21), the frequency domain input signal acquisition step (S30, S31) for the microphone output signal The spectrum analyzer 35 performs the spectrum analysis steps S22 and S23. At this time, if the spectral quality of the acoustic signal input to the microphone 11 is satisfied (i.e., if the acoustic signal has a uniform uniformity of power for each frequency on the spectrum as described above), the update component of the filter coefficient is calculated. If the spectral quality of the sound signal input to the microphone 11 is not satisfied (that is, the uniformity of power for each frequency of the sound signal is less than the preset uniformity). Back side) The process of updating the filter coefficients by calculating the update components of the filter coefficients (S40, S50, S60) is not performed.

In the normal operation environment adaptation step (S5), while adapting to the echo channel 16 as a sound signal input to the microphone during the normal operation of the audio system, the adaptation to the echo channel 16 only when the quality of the sound signal is satisfactory. In order to accurately remove the echo component in response to the echo channel 16 which is changed while using the audio system, the howling rejection performance can be further improved.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, . ≪ / RTI > Accordingly, such modifications are deemed to be within the scope of the present invention, and the scope of the present invention should be determined by the following claims.

10: sound processing means 11: microphone 12: speaker
13 digital part 14 analog part 15 amplifier part
16: echo channel
20: howling removal means 21: filter 22: delay unit
23: subtraction part
30: filter coefficient estimation means 31: filter coefficient updating unit
32: inverse Fourier transform unit 33: multiplier 34: first Fourier transform unit
35 spectrum analysis unit 36 second Fourier transform unit
37: power normalization unit 40: power adjusting means

Claims (11)

In an audio system for outputting a sound signal of the microphone to the speaker,
Howling elimination means for outputting the error signal obtained by subtracting the filtered signal from the acoustic signal of the microphone after filtering the output signal input to the speaker with a filter 21 estimated the reflection channel between the speaker and the microphone ( 20);
Fourier transform the error signal to produce an error signal in the frequency domain, Fourier transform and conjugate the speaker output signal to produce a speaker output signal in the frequency domain, and error in the frequency domain Multiply the signal and the speaker output signal in the frequency domain by the sample order according to the sample order of the signal to obtain the filter coefficient update component in the frequency domain, and then obtain an inverse Fourier transform to obtain the filter coefficient update component in the time domain. Filter coefficient estimation means (30) for updating the coefficient of the filter by adding or subtracting a value obtained by multiplying a predefined update gain by a filter coefficient update component of a time domain from the coefficient of the filter;
Howling removal audio system, characterized in that configured to include a. The method of claim 1,
In the filter coefficient estimating means 30, a Fourier transform is a Fast Fourier Transform (FFT), and an Invers Fourier Transform is an Inverse Fast Fourier Transform (IFFT). Howling removal audio system characterized in that. The method of claim 1,
The howling removal means 20 includes a delay unit 22 for delaying the output signal of the speaker and inputting it to the filter by the time required for the output signal of the speaker to be echoed to the microphone.
The filter coefficient estimating means (30) uses the signal delayed by the delay unit (22) as a speaker output signal. 4. The method according to any one of claims 1 to 3,
The filter coefficient estimating means 30 has a spectrum analyzer 35 for analyzing the frequency component of the error signal, and updates the coefficient of the filter when power for each frequency is dispersed to a predetermined uniformity. Howling elimination audio system. 5. The method of claim 4,
Howling canceling audio system, characterized in that for updating the coefficient of the filter for a predetermined time initially at the power on (ON). 5. The method of claim 4,
The spectral analyzer 35 calculates a degree of power dispersion within a frequency occupying a predetermined upper power range and updates the coefficient of the filter when the uniformity is satisfied. The method according to claim 6,
The uniformity is defined by the ratio value,
The spectrum analyzer 35 may occupy a power average at a frequency occupying a first predefined upper power range, and occupy a second upper power range defined as a next-order power range with respect to the predefined first upper power range. And calculating a coefficient of the filter when the calculated power average differs below the uniformity after calculating the power average at the frequency. 4. The method according to any one of claims 1 to 3,
It is further provided with a power adjusting means 40 for adjusting the power of the sound signal received through the speaker,
The power adjusting means 40,
A power level adjustment filter that gamma curves the input signal and the output signal to make a lower power signal higher and a higher power signal higher;
A peak window filter that smoothes and outputs peaks in an input signal,
Howling removal audio system, characterized in that configured to include a. A modeling method connected to an input and an output of an unidentified system and estimating coefficients of a filter to model an unidentified system,
An error signal obtaining step of obtaining an error signal by subtracting a signal obtained by filtering an input terminal signal of an unidentified system from the output terminal signal of the unidentified system;
A frequency domain error signal obtaining step of Fourier transforming the error signal;
A frequency domain input signal obtaining step of Fourier transforming an input signal input to an unidentified system through an input terminal;
A frequency domain coefficient update component obtaining step of obtaining a frequency domain filter coefficient update component based on the frequency domain input signal and the frequency domain error signal;
A time domain coefficient update component obtaining step of obtaining a time domain filter coefficient update component by inverse Fourier transforming the frequency domain filter coefficient update component;
A filter coefficient updating step of adding or subtracting a value obtained by multiplying a predefined update gain by a time domain filter coefficient update component from coefficients of the filter;
An update persistence determining step of returning to the error signal obtaining step and repeatedly updating a filter coefficient until the preset condition is satisfied;
Unidentified system modeling method characterized in that it comprises a. The method of claim 9,
The frequency domain coefficient update component obtaining step includes:
And a frequency domain filter coefficient update component is obtained by multiplying a frequency domain input signal by a conjugate conjugate signal and a frequency domain error signal in a sample unit according to the sample order of the signal. 11. The method according to claim 9 or 10,
And the input signal in the error signal obtaining step and the coefficient update component obtaining step is a signal obtained by delaying the input terminal input signal by a delay time of the unidentified system.

Images