KR101550501B1 - Residual echo cancellation method and apparatus - Google Patents

Abstract

A residual echo cancellation method and apparatus therefor are disclosed. The residual echo cancellation method includes the steps of: (a) detecting a state of an input signal; (b) determining whether to update a filter weight according to the detected state and removing a first-order echo for the input signal; (c) classifying a far-end signal output through the speaker; And (d) determining whether to estimate a residual echo component using the harmonic component according to the power spectrum density estimation in the first echo canceled signal according to the classification result, thereby removing the residual echo.

Description

[0001] The present invention relates to a residual echo cancellation method and apparatus,

The present invention relates to a residual echo cancellation method and apparatus capable of removing residual echo using a voice harmonic component.

1 is a block diagram of a conventional echo cancellation system.

As shown in FIG. 1, the voice of the user who is re-inputted through the speaker-microphone path of the other party in the communication device in which two persons physically separated from each other interact greatly affects the call quality.

In addition, the necessity of effective elimination of echoes due to the universal expansion of hands-free communication equipment due to the development of mobile communication technology has also increased.

In general, hands-free communication equipment is widely used in automobiles. When a call is made using the hands-free mode call so that the driver can use both hands while driving, the echo cancellation technique is utilized to remove the voice of the other person entering the microphone through the speaker in the car. Another example can be found in the In-Car-Communication (ICC) system to alleviate the driver's cognitive load during conversations between the driver and the person sitting in the rear seat. Unlike the usual conversation situation facing the eyes, both the driver and the passenger are unable to communicate in front of the car. As a result, the driver turns his / her head to the side to talk, which is a harmful factor to safety. In order to prevent this, a microphone-speaker capable of transmitting the voice of the driver to the rear seat is installed. Thus, echo cancellation technology is utilized.

An adaptive filter is used for such echo cancellation. However, since the order of the adaptive filter is shorter than the order of the Room Impulse Response that generates the echo, a perfect echo cancellation is impossible and a residual echo signal remains.

The present invention is to provide a residual echo cancellation method and apparatus capable of accurately measuring and removing residual echo components existing after applying an adaptive filter.

In addition, the present invention is to provide a residual echo cancellation method and apparatus that can effectively remove residual echo using a post-processing filter using characteristics of a far-end signal.

According to an aspect of the present invention, there is provided an apparatus capable of accurately measuring and removing residual echo components present after applying an adaptive filter.

According to an embodiment of the present invention, there is provided an information processing apparatus including: a detection unit detecting a state of an input signal; An adaptive filter unit for determining whether to update a filter weight according to the detected state and removing a first-order echo for the input signal; And an echo cancellation device including a post-processing filter for identifying a harmonic component according to power spectrum density estimation in the first echo canceled signal and estimating and removing residual echo components therefrom.

If the state is a double-talk containing an echo signal and a near-end signal, the adaptive filter unit may use the finally used filter weight without updating the filter weight.

If the state is a Singletalk containing only an echo signal, the adaptive filter unit may continuously update the filter weight.

Wherein the post-processing filter removes the residual echo component from the harmonic component without estimating the residual echo component if the rear-end signal is a non-speech signal, and further comprising a classifier for classifying a far-end signal, can do.

If the rear-end signal is a voice signal, the post-processing filter can estimate and remove the residual echo component using the harmonic component.

According to another aspect of the present invention, there is provided a method of accurately measuring and removing residual echo components present after applying an adaptive filter.

According to an embodiment of the present invention, there is provided a method of detecting an input signal, comprising the steps of: (a) detecting a state of an input signal; (b) determining whether to update a filter weight according to the detected state and removing a first-order echo for the input signal; (c) classifying a far-end signal output through the speaker; And (d) identifying a harmonic component according to the power spectral density estimation in the first echo canceled signal according to the classification result, and estimating and removing residual echo components therefrom.

(D) estimating the residual echo component to remove residual echo if the classification result is a speech signal; And if the classification result is a non-speech signal, removing the residual echo without estimating the residual echo component.

By providing the residual echo removal method and apparatus according to an embodiment of the present invention, residual echo components existing after applying the adaptive filter can be accurately measured and removed.

Accordingly, the present invention can ensure smooth call quality in a hands-free call in an automobile or the like, and has an advantage that a driver's cognitive load using the in-vehicle communication system can be reduced.

1 is a block diagram of a conventional echo cancellation system;
2 is a diagram illustrating a configuration of an echo removal apparatus according to an embodiment of the present invention.
3 is a flow diagram illustrating a method for removing residual echo in accordance with an embodiment of the present invention.
4 is a graph comparing echo cancellation performance results according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention is capable of various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the detailed description. It is to be understood, however, that the invention is not to be limited to the specific embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

The present invention can effectively remove residual echo components after distinguishing speech signals from non-speech signals using the speech harmonic feature by measuring the residual echo component after removing the echo through the adaptive filter.

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

2 is a diagram illustrating the configuration of an echo canceller according to an embodiment of the present invention.

2, an echo removal apparatus 200 according to an embodiment of the present invention includes an input unit 210, a detection unit 215, an adaptive filter unit 220, a classification unit 225, a post- And an output unit 235. [0031]

The input unit 210 is a means for receiving a sound signal from the outside. For example, the input unit 210 may be a microphone.

The detection unit 215 is a unit for detecting the state of an input signal input through the input unit 210. [

For example, a signal input through the input unit 210 such as a microphone may include an echo signal, a near-end voice signal to be transmitted to the other party, and noise.

The detection unit 215 can detect whether a single-talk period including only an echo signal or a double-talk period including an echo signal and a near-end voice signal is analyzed by analyzing the input signal.

At this time, the detector 215 can detect the state of the input signal using the Geigel algorithm. Here, since the Geigel algorithm is obvious to those skilled in the art, a detailed description thereof will be omitted. According to an embodiment of the present invention, it is assumed that the detection unit 215 uses the Geigel algorithm. However, it is natural that the detection unit 215 can be performed using any one of known algorithms for distinguishing the double talk.

The adaptive filter unit 220 determines whether to update the filter weight according to the state of the input signal detected through the detection unit 215, and removes the first-order echo for the input signal.

That is, the adaptive filter unit 220 may estimate the echo component and then remove the estimated echo component from the input signal, and output the filtered echo component to the post-filter unit 230. The adaptive filter unit 220 removes the estimated echo component from the input signal, and does not completely remove the echo component from the input signal. Accordingly, the signal output through the adaptive filter unit 220 includes the residual echo component.

In addition, the adaptive filter unit 220 may consider the state of the input signal in estimating and removing the echo component. That is, if the state of the input signal detected through the detection unit 215 is a single talk period, the adaptive filter unit 220 may continuously update the filter weight to estimate the echo component and remove the echo.

However, if the state of the input signal is a dulk period, the adaptive filter unit 220 does not update the filter weight, estimates the echo component using the last used (i.e., last used) filter weight value, Echo can be removed.

The classifying unit 225 is a means for classifying the far-end signal output through the output unit 235.

For example, the classifying unit 225 can classify a speech signal and a non-speech signal using frequency characteristics of a far-end signal.

In general, a speech signal has a harmonic characteristic in which a pitch component is within a certain range and a relatively large amount of energy exists at a frequency corresponding to an integral multiple of a fundamental frequency. On the other hand, non-speech signals (e.g., car engine sounds, horn sounds, skid sounds, etc.) do not have the features described above. Table 1 below is a table comparing frequency characteristics for human voice signals.

child Woman man Average conversation base wave 265 225 128 Frequency range 208 to 440 155 ~ 334 85-196

The post-processing filter unit 230 estimates the residual echo component using the harmonic component according to the power spectral density estimation in the signal that has been firstly removed through the adaptive filter unit 220 according to the far-end signal classification result of the classification unit 225. [ To remove residual echo.

For example, if the original signal classification result of the classification unit 225 is a speech signal, the post-processing filter unit 230 uses the harmonic component according to the power spectral density estimation in the signal that has been primarily removed through the adaptive filter unit 220 Thereby estimating the residual echo component and removing the residual echo.

A method of estimating the power spectral density in the post-processing filter unit 230 and accurately estimating the residual echo component using the harmonic component in the estimated power spectral density is well known in the art, so a detailed description thereof will be omitted .

However, if the far-end signal classification result of the classification unit 225 is a non-speech signal, the post-processing filter unit 230 uses the harmonic component according to the power spectral density estimation in the signal removed first through the adaptive filter unit 220 Residual echo component can be removed without estimation.

The output unit 235 is means for outputting a sound signal. For example, it may be a speaker.

3 is a flowchart illustrating a method of removing residual echo according to an embodiment of the present invention.

In operation 310, the echo canceller 200 detects an input state of an input signal input through the input unit 210.

For example, the input state of an input signal may be a single talk and a double talk.

In step 315, the echo canceling apparatus 200 determines whether the input state of the input signal is a double talk period.

If the input state of the input signal is a dulk period, the echo canceling apparatus 200 does not update the filter weight of the adaptive filter in step 320, but uses the finally used filter weight (i.e., the last used filter weight value) Thereby estimating the echo component and removing the echo.

However, if the input state of the input signal is a single talk period, the echo canceling apparatus 200 continuously updates the filter weight of the adaptive filter in step 325, and estimates the echo component to remove the echo. For example, the adaptive filter may be an NLMS adaptive filter algorithm. Of course, it is of course also possible to perform using any one of the known adaptive filter algorithms.

A method of updating a filter weight of an adaptive filter and a method of estimating an echo component using an adaptive filter to remove an echo component from an input signal are obvious to those skilled in the art and will not be described.

In operation 330, the echo canceller 200 analyzes and classifies the frequency characteristics of the far-end signal output through the output unit 235.

For example, by analyzing the frequency characteristics of the far-end signal, it is possible to classify whether the far-end signal is a speech signal or a non-speech signal.

Although it is described that the step of analyzing and classifying the frequency characteristics of the far-end signal is performed after 325 in FIG. 3, the step 330 may be performed before the step 310 or may be performed in parallel before the step 320 Do.

In step 335, the echo removal apparatus 200 determines whether the classification result original signal is a speech signal.

If the far-end signal is a speech signal, the echo canceling apparatus 200 estimates the power spectral density in the first-order echo-canceled signal through the adaptive filter in step 340, and outputs the estimated harmonic component in the estimated power spectral density The remaining echo component is estimated by using the residual echo component.

To briefly explain this, a signal input through the input unit 210 includes a near-end speech signal, noise, and echo, respectively. This is expressed by the following equation.

는 근단 음성 신호 성분을 나타내고,

는 노이즈 성분을 나타내며,

는 에코 성분을 나타낸다.here,

Represents a near-end speech signal component,

Represents a noise component,

Represents an echo component.

The echo removal apparatus 200 estimates an echo component through an adaptive filter and subtracts the echo component from the input signal to remove the echo. However, as already mentioned above, this removes the estimated echo from the input signal, so that the echo component can not be precisely removed from the input signal, and the residual echo component is included.

Thus, the signal input to the post-processing filter can be expressed by the following equation (2).

은 잔여 에코 성분을 나타낸다.here,

Represents the residual echo component.

The power spectral density (PSD) of such residual echo components can be predicted by an equivalent exchange function. Assuming that the near-end speech signal, noise, and echo are independent of each other, the equivalent exchange function can be obtained by the number 3.

이고,

는 잔여 에코 성분의 파워 스펙트럼 밀도를 나타내고,

는 마이크를 통해 입력된 입력 신호의 파워 스펙트럼 밀도를 나타내며,

는 제거된 에코 성분의 파워 스펙트럼 밀도를 나타내고,

는 추정된 에코 성분의 파워 스펙트럼 밀도를 나타낸다.here,

ego,

Represents the power spectral density of the residual echo component,

Represents the power spectral density of the input signal input through the microphone,

Represents the power spectral density of the echo component removed,

Represents the power spectral density of the estimated echo component.

The fundamental wave ( F 0) of the residual echo component has the likelihood function of the following equation (3)

는 가중 파라미터를 나타낸다.here,

Represents a weighting parameter.

The harmonic component has integer multiples of fundamental wave.

Hence, the sample value of the predicted residual echo component can be defined as Eq. 5.

Accordingly, the harmonic component of the power spectral density of the residual echo component can be defined by the following equation (6).

이고,

는 콘볼루션 연산자를 나타낸다.here,

ego,

Represents a convolution operator.

For example, if the likelihood function value is greater than the threshold value, the residual echo component is classified as a signal having a harmonic component, and if the value is below a threshold value, it can be classified as a non-speech signal having no harmonic component.

Thus, the echo removal apparatus 200 can be defined as a post-processing filter 7 for estimating and removing the residual echo component using the harmonic component according to the power spectral density estimation of the residual echo signal.

일 수 있다. 여기서, 하모닉 성분의 보상을 위해 잔여 에코 성분의 파워 스펙트럼 밀도는

로 대체될 수 있다. 여기서,

는 혼합 파라미터이다.here,

Lt; / RTI > Here, for compensation of the harmonic component, the power spectral density of the residual echo component is

≪ / RTI > here,

Is a mixing parameter.

As a result of the classification, if the far-end signal is a non-speech signal, the echo canceling apparatus 200 removes the residual echo from the first echo canceled signal through the adaptive filter without estimating the residual echo component according to the power spectral density estimate at step 345.

Accordingly, the echo canceling apparatus 200 according to an embodiment of the present invention can prevent an increase in computation load due to harmonic estimation of a non-speech signal and can prevent degradation in performance due to residual echo estimation reflecting an erroneous harmonic component There is an advantage to be able to.

4 is a graph comparing echo cancellation performance results according to an embodiment of the present invention.

410 denotes an echo cancellation result obtained by applying only the adaptive filter when a voice echo signal is input, 415 denotes an echo cancellation result according to an adaptive filter and a post-processing filter (estimated harmonic residual echo), 420 denotes an adaptive filter and post- And the echo cancellation result according to the filter (harmonic residual echo estimation).

FIG. 4 shows ERLE (Echo Return Loss Enhancement) as an index indicating how much an input echo signal has been removed. That is, the higher the value, the more echo signals are removed.

That is, as shown in FIG. 4, it can be seen that the echo cancellation method according to the embodiment of the present invention can guarantee smooth call quality despite the decrease in SNR.

Meanwhile, the method of removing residual echo component according to an embodiment of the present invention may be implemented in a form of a program command that can be performed through a variety of electronic information processing means and recorded in a storage medium. The storage medium may include program instructions, data files, data structures, and the like, alone or in combination.

Program instructions to be recorded on the storage medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of software. Examples of storage media include magnetic media such as hard disks, floppy disks and magnetic tape, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, magneto-optical media and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as devices for processing information electronically using an interpreter or the like, for example, a high-level language code that can be executed by a computer.

The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention as defined in the appended claims. It will be understood that the invention may be varied and varied without departing from the scope of the invention.

210:
215:
220: adaptive filter unit
225:
230: Post-process filter unit
235:

Claims (7)

A detector for detecting a state of an input signal;
If the detected state is a dubbing period including an echo signal and a near-end signal, the first-order echo for the input signal is removed using the finally-used filter weight without updating the filter weight An adaptive filter unit; And
A post-processing filter for determining a residual echo component estimation using a harmonic component according to power spectrum density estimation in the first echo canceled signal to remove residual echo; And
Further comprising a classifier for classifying a far-end signal,
Wherein the post-processing filter removes the residual echo without estimating the residual echo component using the harmonic component if the rear-end signal is a non-speech signal. delete The method according to claim 1,
If the state is a Singletock period including only an echo signal,
Wherein the adaptive filter unit continuously updates the filter weights. delete The method according to claim 1,
Wherein the post-processing filter estimates and removes a residual echo component using the harmonic component when the rear-end signal is a voice signal. (a) detecting a state of an input signal;
(b) if the detected state is a dubbing period including an echo signal and a near-end signal, the filter weight is not updated, and the first-order echo ;
(c) classifying a far-end signal output through the speaker; And
(d) determining whether to estimate a residual echo component using the harmonic component according to the power spectral density estimation in the first echo canceled signal according to the classification result, and removing the residual echo,
The step (d)
And removing residual echo without estimating a residual echo component using the harmonic component if the classification result is a non-speech signal. The method according to claim 6,
The step (d)
And if the classification result is a speech signal, estimating the residual echo component to remove residual echo.

Images