KR100848789B1 - Postprocessing method for removing cross talk - Google Patents

Abstract

The present invention can significantly reduce the crosstalk signal component included in the separated signal by post-processing using an adaptive filter to further improve the performance of Independent Component Analysis (ICA), which is used as a mixed signal separation algorithm. A post-processing method for removing cross channel interference that may

The present invention provides a separate signal input step of receiving a separated signal by a frequency domain independent element analysis algorithm of a mixed signal, a mixed sound source signal in which a crosstalk signal separated and input through the separated signal input step, and an estimated post-processing sound source. The post-processing sound source channel signal is estimated by multiplying the crosstalk signal estimated in the adaptive filter coefficient updating step from the adaptive filter coefficient updating step of receiving the channel signal and updating the adaptive filter coefficient and the mixed sound source signal mixed with the crosstalk signal. Characterized in that consisting of the interference signal removing step.

Independent element analysis, cross talk signal, interference signal, cross channel

Description

Post-processing method to eliminate crosstalk {POSTPROCESSING METHOD FOR REMOVING CROSS TALK}

1 is a diagram showing separation of two channel signals by a frequency domain independent element analysis algorithm;

2 and 3 is a post-processing configuration diagram using a post-processing method for removing crosstalk from the crosstalk mixed signal separated by the independent element analysis algorithm according to the present invention,

Figure 4 is a comparative example of the result of applying the post-treatment method for removing crosstalk in accordance with the present invention.

The present invention relates to a post-processing method for improving the performance of a mixed signal separation algorithm. In particular, an adaptive filter is used to further improve the performance of Independent Component Analysis (ICA) used as a mixed signal separation algorithm. The present invention relates to a post-processing method for removing cross-channel interference that can significantly reduce crosstalk signal components included in a separated signal by post-processing.

In various fields such as communication, biosignal processing, and voice processing, a technique of separating a signal of each sound source from a signal superimposed with a plurality of sensors, that is, microphones, is very important.

The Blind Source Separation (BSS) method of the sound source signal separation method is that when a mixed signal input from a plurality of microphones is given, even if no prior information other than the number of sound sources included in the mixed signal is provided, input of each microphone is performed. A method of separating the original sound source signal by using the difference between the signals.

The general BSS method performs well in the simulated ideal environment that is built in the lab, but the sound source separation performance is not satisfactory in the real environment. This is because the BSS method assumes that the convolutional mixing filter is a linear finite impulse response filter, which limits the length of the filter. However, because the actual environment violates this assumption, for example, the addition of electrical nonlinear circuit noise in the process of collecting the microphone signal, or the sound sources themselves move.

In order to solve this problem, spectral subtraction is used as a post-processing method to remove the remaining crosstalk signal that is not completely separated by the conventional BSS method. Using the spectral subtraction method, it is possible to efficiently absorb fine inconsistencies between the actual filter and the estimated filter, thereby producing a clean signal free from noise or interference, but the disadvantage of including musical noise due to spectral components below zero. There is this.

There are various methods for BSS. Independent Component Analysis (ICA) has recently been widely used. ICA can be divided into time domain and frequency domain, and considering the calculation amount, it is known that the frequency domain independent factor analysis method is superior. However, even if the signal is separated by the frequency domain ICA, the signal of each channel is not clearly separated, and each crosstalk signal component remains, thereby degrading performance.

The technical problem to be achieved by the present invention is a crosstalk signal included in the separated signal by post-processing using an adaptive filter to further improve the performance of Independent Component Analysis (ICA), which is used as a mixed signal separation algorithm. It is to provide a post-processing method for removing cross-channel interference that can significantly reduce the components.

The present invention provides a separate signal input step of receiving a separated signal by a frequency domain independent element analysis algorithm of a mixed signal, a mixed sound source signal in which a crosstalk signal separated and input through the separated signal input step, and an estimated post-processing sound source. An adaptive filter coefficient updating step of receiving a channel signal and updating the adaptive filter coefficient and a mixed sound source signal mixed with the crosstalk signal are multiplied by the crosstalk signal estimated in the adaptive filter coefficient updating step to obtain a post-processing sound source channel signal. And estimating an interference signal removing step.

In addition, the separated signal by the independent element analysis algorithm, which has passed through the separated signal input step, is represented by an equation of a frequency domain as shown in Equation 1 below.

<Equation 1>

(Where i is each channel, w is frequency, t is time, (s) is the desired source channel signal, and (c) is crosstalk signal).

In the adaptive filter coefficient updating step, the adaptive filter coefficient may be updated by the following equation.

<Equation 2>

은 시간 인덱스이며,

는 적응필터의 입력신호의 크기,

는 혼합음원신호와 추정 크로스토크신호의 오차신호 크기,

는 적응상수이며 ,

는 적응계수부의 발산을 막기 위한 미소상수이다.)(here,

Is the time index,

Is the magnitude of the input signal of the adaptive filter,

Is the magnitude of the error signal between the mixed sound source signal and the estimated crosstalk signal,

Is the adaptive constant,

Is a small constant to prevent the divergence of the adaptive coefficients from diverging.)

Further, in the interference signal removing step, the interference signal may be removed by the following equation.

<Equation 3>

는 혼합음원신호 그리고

는 추정 크로스토크신호를 나타낸다.)(here,

Is the mixed sound source signal and

Denotes an estimated crosstalk signal.)

Another aspect of the present invention is a sound source separation unit for separating the signal by the frequency domain independent element analysis algorithm for the input signal, the mixed sound source signal mixed with the input crosstalk signal separated from the sound source separation unit and estimated post-processing The post-processing sound source channel signal is obtained by multiplying the crosstalk signal estimated in the adaptive filter coefficient updating step from the adaptive filter coefficient updater for receiving the sound source channel signal and updating the adaptive filter coefficient and the mixed sound source signal mixed with the crosstalk signal. It is to provide a post-processing device for removing crosstalk, characterized in that it comprises an interference signal cancellation unit for estimating.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings.

1 is a diagram illustrating the separation of two channel signals by a frequency domain independent element analysis algorithm, and FIGS. 2 and 3 remove crosstalk signals from the crosstalk mixed signal separated by the independent element analysis algorithm according to the present invention. 4 is a comparative example of the results of applying the post-processing method for removing the crosstalk signal according to the present invention.

,

는 각각 원신호(source signal) 들이고

,

는 두 원신호가 특정한 전달경로에 의하여 혼합된 신호이다.

,

는 독립요소분석에 의하여 분리된 결과 신호이다. 이들 신호 속에 각각의 크로스토크 성분 신호들이 존재하게 된다. 즉, 원신호

의 신호성분이

에 포함되어 있고

에

의 성분이 존재하여 신호가 완전히 분리되지 않게 된다. 이를 주파수 영역의 식으로 Referring to FIG. 1, a schematic diagram of a system for separating two channel signals in the separated signal input step is displayed based on a frequency domain independent element analysis method among independent element analysis methods. Shown in FIG.

,

Are the source signals, respectively

,

Is a signal in which two original signals are mixed by a specific transmission path.

,

Is the result signal separated by independent factor analysis. In these signals, respective crosstalk component signals are present. That is, the original signal

Signal component of

Included in

on

The presence of the component prevents the signal from being completely separated. This is expressed in the frequency domain

는 각 채널을 나타내는 인자,

는 주파수,

는 시간,

는 원하는 음원채널 신호 그리고

는 크로스 토크 신호이다.)로 나타낸다.(here,

Is a factor representing each channel,

Is frequency,

Time,

Is the desired source channel signal and

Is a crosstalk signal.

는

속에 포함되어 있는

성분을 제거하기 위한 적응필터의 계수이고

는

에 포함되어 있는

성분을 제거하는 적응계수이다. 적응필터는In addition, after the adaptive filter coefficient updating step and the interference signal removing step using the post-processing method for removing the crosstalk signal from the crosstalk mixed signal separated by the independent element analysis algorithm according to the present invention of FIGS. It demonstrates with reference to a process block diagram. First of all,

Is

Contained in

Coefficients of the adaptive filter to remove the components

Is

Included in

Adaptive factor to remove components. Adaptive filter

은 시간 인덱스이며,

는 적응필터의 입력신호의 크기,

는 혼합음원신호와 추정 크로스토크신호의 오차신호 크기,

는 적응상수이며 ,

는 적응계수부의 발산을 막기 위한 미소상수이다.)로 나타내어 진다. 위 식에 의하여 각 주파수 빈에 대한 각각의 적응필터에 의하여 적응계수를 산출하게 된다. 도 3에 나타난 바와 같이, 음원분리부는 입력신호에 대하여 주파수영역 독립요소분석알고리즘에 의해 신호를 분리하며, 적응필터계수갱신부는 상기 음원분리부로부터 분리되어 입력된 크로스토크신호가 혼합된 혼합음원신호와 추정된 후처리음원채널신호를 입력받아 적응필터 계수를 갱신하고, 상기 간섭신호제거부는 크로스토크신호가 혼합된 혼합음원신호에서 상기 적응필터계수갱신단계에서 추정된 크로스토크신호를 승산하여 후처리음원채널신호를 추정한다.(here,

Is the time index,

Is the magnitude of the input signal of the adaptive filter,

Is the magnitude of the error signal between the mixed sound source signal and the estimated crosstalk signal,

Is the adaptive constant,

Is a microconstant to prevent divergence of the adaptive coefficient part. According to the above equation, the adaptive coefficient is calculated by each adaptive filter for each frequency bin. As shown in FIG. 3, the sound source separation unit separates the signal from the input signal by a frequency domain independent element analysis algorithm, and the adaptive filter coefficient update unit is separated from the sound source separation unit and mixed with the input crosstalk signal. And after receiving the estimated post-processing sound source channel signal, the adaptive filter coefficient is updated, and the interference signal removing unit multiplies the crosstalk signal estimated in the adaptive filter coefficient updating step from the mixed sound source signal mixed with the crosstalk signal. Estimate the sound source channel signal.

In the interference signal removing step,

는 혼합음원신호 그리고

는 추정 크로스토크신호를 나타낸다.)에 의하여, 간섭신호를 제거한다. 본 발명은 오차신호의 크기로 정규화함으로써

이 오조정되는 것을 방지할 수 있다.(here,

Is the mixed sound source signal and

Denotes an estimated crosstalk signal). The present invention is normalized to the magnitude of the error signal

This misalignment can be prevented.

의 크기 와 오차신호

의 크기로 적응된다. 적응필터가 동작하여 오차신호의 크기가 적어질 경우 수학식 2의 우측 2번째 항인

은 0에 가깝게 수렴 되어

신호로부터 크로스토크 성분의 효과적 제거가 가능하게 된다. According to equations (2) and (3), the update of the adaptive coefficient is an input signal.

Magnitude and error signal

Is adapted to the size of. When the adaptive filter operates to reduce the magnitude of the error signal, the second term on the right side of Equation 2

Converges close to zero

It is possible to effectively remove the crosstalk component from the signal.

In addition, the effect can be confirmed through a comparative example of the result of applying the post-processing method for removing the crosstalk signal according to the present invention of FIG. Set the speaker at 12cm and place the speaker 100cm away from the center between the two microphones. A male and female voice, quantized in 16 bits and sampled at 16 kHz, was output to each speaker and separated. The conversion to the frequency domain was 1024 samples, 64 samples were overlapped and the Hamming window was used. The number of taps of the adaptive filter for post processing is 16 bits and the adaptation coefficient is 0.2. The signals are transformed into the frequency domain and shown for the 100 Hz frequency bin.

Among the results, (a) and (c) are signals separated by frequency domain independent element analysis, respectively. The presence of crosstalk signal components in each channel is indicated by circles. (b) and (d) are the result of the post-processing of the present invention and it can be seen that the crosstalk signal component is significantly reduced.

The present invention can be widely applied to various fields such as improving the recognition performance of a voice recognition system, improving the voice quality of a voice communication system such as a mobile phone and a hearing aid.

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, which are also implemented in the form of carrier waves (for example, transmission over the Internet). It also includes. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. And functional programs, codes and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.

As described above, according to the present invention, the crosstalk signal component remaining in the resultant signal after the signal separation by the independent element analysis algorithm is significantly reduced by using an adaptive filter, thereby increasing the effect of the independent element analysis algorithm. It can be applied not only in stationary noise environment but also in non-stationary noise environment, so that in various fields such as communication, biosignal processing, and voice processing, a plurality of sensors, that is, microphones, are recorded and overlapped. Many effects can be expected by applying the present invention to various industrial fields that separate the signal of each sound source from the received signal.

Although the present invention has been described with reference to the above embodiments, it is merely illustrative, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. . Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

Claims (5)

delete A separation signal input step of receiving a separation signal by a frequency domain independent element analysis algorithm of the mixed signal; An adaptive filter coefficient updating step of receiving a mixed sound source signal mixed with the input crosstalk signal separated through the separated signal input step and an estimated post-processing sound source channel signal and updating an adaptive filter coefficient; And And an interference signal removing step of estimating a post-processing sound source channel signal by multiplying the crosstalk signal estimated in the adaptive filter coefficient updating step from the mixed sound source signal mixed with the crosstalk signal. And a separation signal by the independent element analysis algorithm, which has passed through the separation signal input step, is represented by a frequency domain equation as shown in Equation 1 below. <Equation 1>

(Where i is each channel, w is frequency, t is time, (s) is the desired source channel signal, and (c) is crosstalk signal). The post-processing method of claim 2, wherein in the adaptive filter coefficient updating step, the adaptive filter coefficient is updated by the following equation. <Equation 2>

(here,

Is the time index,

Is the magnitude of the input signal of the adaptive filter,

Is the magnitude of the error signal between the mixed sound source signal and the estimated crosstalk signal,

Is the adaptive constant,

Is a small constant to prevent the divergence of the adaptive coefficients from diverging.) The post-processing method of claim 3, wherein in the interference signal removing step, the interference signal is removed by the following equation. <Equation 3>

(here,

Is the mixed sound source signal and

Denotes an estimated crosstalk signal.) delete

Images