KR20070085193A - Noise cancellation apparatus and method thereof - Google Patents

Abstract

A noise removing device and a method therefor are provided to separate a frequency response of a target signal and a frequency response of a noise signal by using position information of the target signal and remove a noise. A plurality of microphones(100a,100b) receives sound signals in which a voice and a noise are included. A target signal information input unit(200) receives a position of a target signal, which is a voice signal, as preceding information. An equalizer information input unit(300) outputs equalizer information for correcting mismatching among a plurality of the microphones(100a,100b). A noise removing unit(400) analyzes frequencies of the sound signals, and estimates whether respective frequency responses of the sound signals are a frequency response of a noise signal or a frequency response of the target signal on the basis of the preceding information. A control unit(500) controls the operation of a plurality of the microphones(100a,100b), the target signal information input unit(200), the equalizer information input unit(300), and the noise removing unit(400). The control unit(500) analyzes the frequencies of the sound signals received through a plurality of the microphones(100a,100b), estimates whether the respective frequency responses are the frequency response of the noise signal or the frequency response of the target signal on the basis of the preceding information, and removes a noise by using the estimated result.

Description

Noise reduction device and method {NOISE CANCELLATION APPARATUS AND METHOD THEREOF}

The present invention relates to a noise canceling device and method for speech recognition and speech decoder. In particular, the present invention relates to a method for removing noise by separating the frequency response of the target signal and the frequency response of the noise signal using the location information of the target signal.

Noise, which is an undesirable signal in speech recognition systems, is divided into normal noise and abnormal noise.

As a method for removing the normal noise, a Wiener filter method and a Kalman filter method are generally used.

The Wiener filter method estimates the frequency response of the noise signal and detects the voice signal by subtracting the frequency response of the noise signal estimated from the frequency response of the arbitrary input signal. In other words, the Wiener filter method starts by estimating the frequency response of the noise signal at the initial 100msec ~ 200msec of the input signal or continuously estimating the non-voice interval using a real-time voice detector. Then, if any input signal comes in, the frequency response is obtained and the power spectrum of the speech signal to be detected is obtained by subtracting the power spectrum of the estimated noise. Through this, both the noise and the frequency response of speech can be found, and the system function for noise cancellation can be estimated. The sample value itself of the speech signal to be detected can be obtained through the process of convolution of the input signal and the system function in the time domain. Equation 1 calculates a power spectrum of a voice signal to be detected from a given input signal.

[Equation 1]

는 검출하고자 하는 음성 신호의 주파수 응답,

는 잡음이 섞인 입력 신호의 주파수 응답을 나타내고,

는 입력 신호의 초기 신호 혹은 실시간 음성 검출기를 통해서 추정된 잡음 신호의 주파수 응답이다. here,

Is the frequency response of the voice signal to be detected,

Represents the frequency response of the noisy input signal,

Is the frequency response of the noise signal estimated by the initial signal of the input signal or the real-time voice detector.

Equation 2 is a design method of the Wiener filter.

[Equation 2]

Equation 3 is a noise removal method in the time domain.

[Equation 3]

은 잡음이 제거된 음성 신호, '

'는 선형복적분 기호이고, IFFT(Inverse Fourier Transform)은 역푸리에 변환 기호이다. here,

Is a noise-free speech signal,

'Is the linear double sign and IFFT (Inverse Fourier Transform) is the inverse Fourier transform symbol.

The Kalman filter method statistically analyzes the value of the optimal speech signal obtainable from a given input up to time n using the optimal speech signal value and optimal Kalman gain obtained from the given input up to time (n-1). Say the way. In addition, unlike the Wiener filter method, which is performed only in the pure time domain and is a non-casual system, it is a causal system.

Equation 4 is a noise reduction method by the Kalman filter.

[Equation 4]

은 시간 n까지 주어진 입력으로부터 구할 수 있는 최적 음성 신호 값,

는 (n-1) 까지 주어진 입력으로부터 구할 수 있는 최적 음성 신호의 값,

은 최적 칼만 이득,

는 음성 신호의 생성에 직접 관련이 있는 선형 예측 계수이다.here,

Is the optimal speech signal value from a given input up to time n,

Is the value of the optimal speech signal that can be obtained from a given input up to (n-1),

Optimum Kalman benefit,

Is a linear prediction coefficient that is directly related to the generation of the speech signal.

However, this Wiener and Kalman filter method is an effective algorithm when the mixed noise is a normal noise. In other words, if the object to be removed is abnormal, such as music or human voice, the effect is not achieved.

As a method for removing abnormal noise, a blind source seperation method and a beamforming method may be generally used. These methods use a microphone array, and when the microphone array is used, location information and path information of a target signal to be detected may be used.

The blind signal separation method is an algorithm used when the number of microphones is equal to or greater than the number of mixed signal sources including noise and the target signal. The representative algorithm is an independent component algorithm (ICA). In the ICA technique, the system function between the signal source and the microphone is given by the path function, and the inverse of the path matrix is obtained under the assumption that the signal sources are non-Gaussian. That is, the inverse matrix is trained so that the mutual information of the microphone array signals passing through the inverse of the path matrix is maximized.

However, the blind signal separation method has a limitation that the number of signal sources should always be smaller than the microphone. The number of signal sources cannot be determined quantitatively, and in practical cases, they are scattered over a large area, so good performance cannot be expected. In addition, since the blind signal separation method takes a long time for the algorithm to converge in the signal separation process, it cannot be used basically for signal detection for speech recognition, and likewise, it is difficult to easily apply to a decoder for voice communication.

The beamforming technique uses the position of the target signal to be detected as the preceding information unlike the blind signal separation technique. In beamforming, there is no limit on the number of microphones. The more microphones, the better the performance. Recently, the most widely used beamforming technique is a generalized sidelobe canceller (GSC) technique. The GSC technique performs fixed beamforming based on the position of the target signal input as the preceding information and steers the microphone array to increase the signal-to-noise ratio in the direction. The microphone array signals are then phase matched in the direction of the target signal to obtain a signal with a high signal-to-noise ratio. However, the noise signal still remains, and an adaptive filtering technique using the reference noise signal is applied to remove the residual noise that exists after the fixed beamforming.

1 is an adaptive filter based noise cancellation technique using a reference noise signal.

는 고정 빔포밍 이후의 신호이고,

는 고정 빔포밍 출력에 섞여 있는 잡음

을 제거하기 위한 참조 잡음 신호를 나타낸다. 참조 잡음 신호는 선행 입력된 목적 신호의 위치 정보를 바탕으로 마이크 어레이의 채널간 신호를 적절히 차감하는 방법으로 추정될 수 있다. In Figure 1

Is the signal after fixed beamforming,

Is noise mixed in a fixed beamforming output

Represents a reference noise signal to remove. The reference noise signal may be estimated by a method of properly subtracting the inter-channel signal of the microphone array based on the position information of the previously input target signal.

Beamforming techniques are commonly used in radiocommunication communication systems and use a powerful adaptive filter to remove noise. However, in the beamforming technique for processing a speech signal, an adaptive filter removes noise and severely distorts speech. This is because the signal characteristics of the noise to be removed are similar to those of the target signal. Accordingly, there is a problem in that a sound quality is deteriorated in the decoder for voice communication, and the recognition rate is lowered due to distortion of the frequency response when used in preprocessing for voice recognition.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide an apparatus and a method for removing noise for a speech recognition and speech decoder that can improve speech recognition rate and improve call quality.

According to one aspect of the present invention, a noise canceling method is provided. A noise reduction method for speech recognition and speech decoder includes: (a) receiving a position of a target signal which is a speech signal as preceding information; (b) after frequency analysis of the multi-channel acoustic signals received from the plurality of microphones, on the basis of the preceding information, it is estimated whether each frequency response of the received signal is a frequency response of a noise signal or a frequency response of a target signal. Doing; And (c) performing noise cancellation using the frequency response of the noise signal estimated in step (b) and the frequency response of the target signal.

According to another feature of the present invention, a noise canceling device is provided. A noise canceling device for speech recognition and speech decoder includes: a plurality of microphones for receiving an audio signal including speech and noise; An object signal information input unit which receives a position of the object signal as a voice signal as preceding information; An equalizer information input unit for outputting equalizer information for correcting mismatches between the plurality of microphones; A noise canceling unit for frequency analyzing the acoustic signal and estimating whether each frequency response of the received acoustic signal is a frequency response of a noise signal or a frequency response of a target signal based on the preceding information; And a control unit controlling operations of the plurality of microphones, the target signal information input unit, the equalizer information input unit, and the noise canceling unit.

According to the present invention, it is possible to provide an apparatus and a method for removing noise for speech recognition and a speech decoder which can improve speech recognition rate and improve call quality.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is said to "include" a certain component, it means that it can further include other components, except to exclude other components unless otherwise stated. In addition, the terms “… unit”, “… unit”, “module”, etc. described in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware or software or a combination of hardware and software. have.

Hereinafter, an apparatus and a method for removing noise according to an embodiment of the present invention will be described in detail.

2 is a schematic block diagram of an apparatus for removing noise according to an embodiment of the present invention, and FIG. 3 is a block diagram of a noise removing unit according to an embodiment of the present invention.

Noise reduction apparatus according to the embodiment of the present invention includes a plurality of microphone (100a, 100b), the target signal information input unit 200, the equalizer information input unit 300, the noise canceling unit 400 and the control unit 500. . At this time, each configuration is briefly described as follows.

The microphones 100a and 100b receive sound generated from the outside and convert the sound into a corresponding sound signal. At this time, the acoustic signal includes a desirable voice signal and an undesirable noise signal. The microphone is composed of a plurality of microphones, only two of them are shown here for convenience of description.

The target signal information input unit 200 receives a position of the target signal to be acquired and transmits it to the noise canceller 400.

The equalizer information input unit 300 transmits equalizer information to the noise canceller 400 to correct mismatch between microphones.

The noise remover 400 extracts the frequency response of the target signal and the frequency response of the noise signal from the received acoustic signal based on the positional information and the equalizer information of the target signal, and detects the target signal from which the noise is removed.

The controller 500 controls operations of the microphones 100a and 100b, the object signal information input unit 200, the equalizer information input unit 300, and the noise canceling unit 400.

A more detailed configuration of the noise removing unit 400 will be described with reference to FIG. 3.

3 is a block diagram showing a detailed configuration of the noise removing unit 400 according to an embodiment of the present invention.

Referring to FIG. 3, the noise remover 400 includes a signal block forming unit 410, a frequency divider 420, a frequency response estimator 430, a noise processor 440, a frequency synthesizer 450, and a signal. Termination determination unit 460 is included.

The signal block forming unit 410 divides the input sound signal into sound signals in unit time units. That is, the sound signal is divided into short-term multi-channel signal blocks.

The frequency dividing unit 420 performs frequency division filtering in the time domain to solve the ambiguity of the time delay and the direction of the multi-channel time block input for each unit time.

The frequency response estimator 430 performs channel-specific frequency analysis on the channel-frequency band signal, and estimates direction information on the signal block of each frequency band. Thereafter, the frequency response estimator 430 estimates the frequency response of the noise and the target signal according to the estimated direction information. However, since there is an undefined part of the estimated frequency response, the total frequency response is estimated through frequency response smoothing.

The noise processor 440 removes noise using the Wiener filter coefficients obtained in the time domain from the frequency response of the entire target signal and the noise signal estimated by the frequency response estimator 430. Through this, the target signal for each frequency band is extracted.

The frequency synthesizer 450 performs time domain frequency synthesis filtering to synthesize signals of the entire frequency band from the target signal for each frequency band extracted by the noise processor 440.

The signal termination determination unit 460 determines whether the input sound signal ends, and if the signal is not terminated, detects the target signal for the next signal block.

Now, a noise cancellation method according to an embodiment of the present invention will be described.

4 is an overall flowchart of a noise canceling method according to an exemplary embodiment of the present invention.

The noise canceling apparatus of the present invention receives the position information of the target signal (S101), and receives the equalizer information (S102) for correcting mismatch between microphones.

Thereafter, a process of extracting a target signal with respect to the multi-channel signal block input for each unit time is performed.

First, a short-term multi-channel signal is input (S103), and time domain frequency division filtering to solve ambiguity for time delay and direction is performed (S104) to generate a channel-frequency band signal.

Thereafter, channel-specific frequency analysis is performed on the channel-frequency band signal (S105), and direction information is estimated based on the equalizer information (S106). Thereafter, the short-term frequency response of the noise and the target signal is estimated according to the estimated direction information (S107). At this time, the estimation of the frequency response of the noise signal and the target signal is estimated by a method of determining whether each frequency is the noise signal or the target signal based on the target signal position information.

Since there is an undefined portion of the frequency response of the estimated noise and the target signal, the overall frequency response is estimated through frequency response smoothing (S108). Thereafter, noise removal is performed using the Wiener filter coefficients obtained over time (S109).

Through the above process, the target signal for each frequency band is extracted, and time domain frequency shamble filtering is performed to synthesize a signal of the entire frequency band (S110).

Subsequently, it is determined whether the signal is terminated (S111), and if the signal is not terminated, the target signal is detected for the next signal block.

Hereinafter, each step will be described in detail.

In the method of removing noise according to an embodiment of the present invention, position information of a target signal is first input.

In the step of inputting position information of the target signal (S101), the position of the target signal to be acquired is received. When the distance between the target signal to be acquired and the terminal receiving the target signal is short distance of 1 m or less, the component by the direct path is large when the acoustic signal enters the microphone, so that the reverberation component can be almost ignored. If the reverberation is weak, when two or more microphones are used, it is possible to detect at which angle each frequency component is received without significant error.

Therefore, when using the position of the target signal as the preceding information as in the present embodiment, a training process for identifying the position of the target signal is not necessary. That is, even when a signal is received, the noise component and the target signal component can be separated in the frequency domain in real time using the location information of the target signal.

Subsequently, the noise canceling method according to the present embodiment passes through an equalizer information input step S102. In the equalizer information input step S102, mismatch between microphones is corrected. In the voice recognition environment, the difference between the characteristics of the microphones is inevitable, and the characteristics of the analog-to-digital (A / D) converter that receives the signal from the microphone may also be different. The equalizer information attraction step S102 compensates for the difference in characteristics of the microphone and the A / D converter in the frequency domain.

Equation 5 shows a cost function for the equalizer implementation when the reference channel is number 1 and the number of input channels is two.

[Equation 5]

는 시간, T는 총 음성 신호 블록의 개수이다. Where k is the discrete frequency,

Is the time, and T is the total number of speech signal blocks.

이며, FFT[]는 고속 푸리에 변환 함수이다.

과

은 채널별 입력 신호이다.

는 채널 2의 주파수 응답을 보정하기 위한 등화기 계수이다. 수학식 5을 최적화하면, 수학식 6를 얻는다.

FFT [] is a fast Fourier transform function.

and

Is an input signal for each channel.

Is the equalizer coefficient for correcting the frequency response of channel 2. By optimizing Equation 5, Equation 6 is obtained.

[Equation 6]

5 is a flowchart of an equalizer design method required for an embodiment of the present invention.

First, the positional information of the target signal is input (S201), and the short-term multi-channel signal is input for each unit time (S202). Thereafter, it is determined whether the input signal is a training signal (S203). If the input signal is not a training signal, the signal is received per unit time again.

If the input signal is a training signal, frequency analysis for each channel is performed (S204), power is accumulated for each frequency of the reference channel (S205), and frequency correlation between channels is accumulated (S206).

Thereafter, it is determined whether the signal input is terminated (S207), and if the signal input is not finished, the signal for each unit time is input again.

When the signal is input, the ratio of the cumulative correlation and the cumulative power is calculated (S208).

Thereafter, a process of extracting a target signal with respect to the multi-channel signal block input for each unit time is performed. A short-term multi-channel signal is input (S103), and time-domain frequency division filtering is performed to solve ambiguity of time delay and direction (S104) to generate a channel-frequency band signal.

In the time-domain frequency division filtering step S104, the signal is divided by performing equal band division filtering and down-sampling in the frequency domain. 6 is a flowchart illustrating a time domain frequency division filtering method.

/M ) 통과 신호로 분할한다. 그러면, (1/M)의 신호로 분할된다(S303). Referring to FIG. 6, the apparatus for removing noise receives an input of a short-term signal (S301) and performs band pass filtering at equal intervals (S302). Entire signal in M equally spaced bands (

/ M) Split into pass signals. Then, the signal is divided into a signal of (1 / M) (S303).

In the channel-specific frequency analysis (S105), frequency analysis for each channel is performed on the divided signal. That is, unit short-term frequency analysis is performed on the signal of the channel-frequency band.

In the direction information estimation step (S106), the direction information of each frequency component is estimated using the frequency response for each channel. The direction information can be estimated using the gain difference between channels and the time difference between channels for each frequency component.

For example, when the number of input channels is two, it can be obtained through the process of Equation 7, Equation 8 and Equation 9.

[Equation 7]

[Equation 8]

[Equation 9]

는 채널간 이득차,

는 채널간 시간차,

은 FFT의 크기를 나타낸다. here,

Is the gain difference between channels,

Is the time difference between channels,

Represents the size of the FFT.

In the step of frequency response estimation (S107), classification is performed for each unit frequency response of the target signal and the noise based on the position information of the previously input target signal. Assuming that the position of the target signal is located in front of the microphone array can be performed as shown in Equation 10.

[Equation 10]

값은 (채널간 이득차, 채널간 시간차)로 주어지는 평면에서 목적 신호를 추출하기 위한 문턱치 값이다. 수학식 6에서

이 1일 경우에는 기준 채널의 k 번째 단위 구간 주파수 응답이 목적 신호로부터 나온 것으로, 0일 경우에는 잡음원으로부터 나온 것으로 추정한다.

The value is a threshold value for extracting the target signal in the plane given by (inter-channel gain difference, inter-channel time difference). In equation (6)

If it is 1, the k-th unit frequency response of the reference channel is derived from the target signal, and if it is 0, it is assumed to be from a noise source.

In the step of frequency response smoothing (S108), an undefined frequency response in the noise and the target signal is estimated. That is, since there is an undefined part of the frequency response of the noise and the target signal estimated in the previous step, the total frequency response is estimated through frequency response smoothing. (11) shows the method.

[Equation 11]

는 스무싱된 주파수 응답 크기,

은 입력된 주파수 응답 크기를 나타내고,

는 스무싱된 함수의 주파수 응답 크기를 나타낸다. here,

Is the smoothed frequency response magnitude,

Represents the input frequency response magnitude,

Denotes the magnitude of the frequency response of the smoothed function.

Thereafter, noise reduction is performed using coefficients of the Wiener filter obtained in the time domain in the entire frequency response (S109).

Through the above steps, the target signal for each frequency band for the input signal is extracted, and then time domain frequency synthesis filtering (S110) is performed to synthesize the signals of the entire frequency band.

,

)에서만 정의되어야 한다. 이것은 디지털 주파수

성분의 신호가 가질 수 있는 시간차가 1 샘플 미만이어야 한다는 뜻과 같다. In the step of time domain frequency synthesis filtering (S110), up-sampling and equal band synthesis filtering of signals for each band are performed. The time-domain frequency division filtering for each channel is necessary because the ambiguity in the signal source can be eliminated. To increase the sampling frequency and eliminate spatial ambiguity, the distance between the microphones must be reduced. Phenomenologically, the phase response used in Equation 9 to calculate the time difference between channels is (-

,

Must be defined only). This is digital frequency

Equivalent to the time difference that a component's signal can have.

,

)가 되도록 하는 조건을 항상 만족시킬 수 있다. 도 7은 분할할 균등 주파수 밴드가 2개일 때 다중 위상 QMF(quardrature mirror filtering)기법으로 구현한 예시이다. 도 7에서

,

은 저대역 및 고대역 분할 신호이며,

는 입력

와 주파수 영역에서 전력만이 보존된 신호이다. 즉, 위상 응답은 다를 수 있다. According to this embodiment, the sampling frequency can be lowered as the number of equal frequency bands is increased, so that the phase response of Equation (9) is given for the interval between microphones.

,

Can always be satisfied. FIG. 7 illustrates an example of multi-phase QMF (quadrature mirror filtering) when two equal frequency bands are divided. In Figure 7

,

Is a low band and high band split signal,

Enter

Only power in the and frequency domains is conserved. That is, the phase response can be different.

8 is a flowchart illustrating a time domain frequency synthesis filtering method.

/M ) 통과 필터링을 수행(S403)하고, 각 대역 신호를 합하여(S404), 전체 주파수 대역의 신호를 합성한다. M band pass signals are input (S401), and M times zero for each band signal is inserted (S402). Subsequently, for each band signal, an equal interval band (

/ M) Pass filtering is performed (S403), and each band signal is summed (S404) to synthesize signals of all frequency bands.

 Thereafter, in step S111, if the signal is not terminated, the target signal is detected for the next signal block, and if the signal is terminated, the voice recognition method ends.

EXAMPLE

FIG. 9 illustrates a gain difference and time difference distribution of frequency components between channels in an embodiment. In the embodiment, the sampling frequency is 16 kHz, the number of microphones is 2, the interval between microphones is 4 cm, the number of equal band divisions is 2, the size of the FFT is 512, the target signal source and the two noise signal sources, and the angle between the noise signal sources is 150 FIG. 6 shows the time difference and gain difference between channels for the low band and high band signals when r is 0.4 and SNR is 0 dB. Referring to FIG. 9, it can be seen that the target signal component and the noise signal component are combined, and FIG. 10 shows a waveform of a signal mixed with noise and a result of noise removal according to the present embodiment.

According to the present exemplary embodiment, the frequency response of the target signal and the frequency response of the noise signal are separated based on the preceding position information, and the noise is removed by using the frequency response of the target signal and the noise signal. Through this, it is possible to provide a noise reduction device and method for speech recognition and speech decoder which can improve speech recognition rate and improve call quality.

The embodiments of the present invention described above are not implemented only through the apparatus and the method, but may be implemented through a program for realizing a function corresponding to the configuration of the embodiment of the present invention or a recording medium on which the program is recorded. Implementation may be easily implemented by those skilled in the art from the description of the above-described embodiments.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

1 is an adaptive filter based noise cancellation technique using a reference noise signal.

2 is a schematic block diagram of an apparatus for removing noise according to an embodiment of the present invention, and FIG. 3 is a block diagram of a noise removing unit according to an embodiment of the present invention.

4 is an overall flowchart of a noise canceling method according to an exemplary embodiment of the present invention.

5 is a flowchart of an equalizer design method required for an embodiment of the present invention.

6 is a flowchart illustrating a time domain frequency division filtering method.

FIG. 7 illustrates an example of multi-phase QMF (quadrature mirror filtering) when two equal frequency bands are divided.

8 is a flowchart illustrating a time domain frequency synthesis filtering method.

9 illustrates a distribution of gain and time differences in frequency components between channels according to an embodiment.

10 illustrates a signal before and after noise cancellation according to an embodiment.

Claims (9)

In the noise reduction method for speech recognition and speech decoder, (a) receiving a position of a target signal as a voice signal as preceding information; (b) after frequency analysis of the multi-channel acoustic signals received from the plurality of microphones, the frequency response of the target signal is whether each frequency response of the signal received in the step (a) is a frequency response of the noise signal based on the preceding information; Estimating whether it is a response; And (c) performing noise cancellation using the results of step (b) Noise Reduction Method. The method of claim 1, In step (b), (i) performing time domain frequency division filtering on the received multichannel acoustic signal; (ii) performing channel-specific frequency analysis on the signal frequency-divided in step (i); (iii) estimating direction information based on the frequency analysis of step (ii); And (iv) estimating the short-term frequency response of the target signal and the noise signal based on the direction information estimated in step (iii); Noise Reduction Method. The method of claim 2, In step (c), (v) estimating the total frequency response of the noise signal and the target signal using the frequency response of the noise signal estimated in step (b) and the frequency response of the previously defined portion of the undefined part of the frequency response of the target signal Making; And (vi) removing noise from the overall frequency response using coefficients of the Wiener filter obtained in the time domain; Noise Reduction Method. The method of claim 3, In step (a), (vii) receiving frequency domain equalizer information using a gain difference and a time difference between channels to compensate for interchannel characteristics of the multichannel speech signal; Noise Reduction Method. The method of claim 4, wherein Step (vii), (1) determining whether the received multichannel sound signal is a training signal; (2) if the received multi-channel sound signal is a training signal in step (1), performing frequency analysis for each channel; (3) accumulating power for each frequency of the reference channel and accumulating correlation for each frequency between the channels; (4) determining whether the input of the training signal is finished; And (5) calculating the ratio of cumulative correlation and cumulative power when the input of the training signal in step (4) is terminated; Noise Reduction Method. The method of claim 5, In step (i), the received multi-channel sound signal may be divided into M equally spaced bands (

/ M) divided by the pass signal Noise Reduction Method. The method of claim 6, In step (iv), the short-term frequency response of the target signal and the noise signal are estimated by using a gain difference and a time difference between frequency components. Noise Reduction Method. In the noise reduction device for speech recognition and speech decoder, A plurality of microphones for receiving an acoustic signal including voice and noise; An object signal information input unit which receives a position of the object signal as a voice signal as preceding information; An equalizer information input unit for outputting equalizer information for correcting mismatches between the plurality of microphones; A noise canceling unit for frequency analyzing the acoustic signal and estimating whether each frequency response of the received acoustic signal is a frequency response of a noise signal or a frequency response of a target signal based on the preceding information; And It includes a control unit for controlling the operation of the plurality of microphones, the target signal information input unit, the equalizer information input unit and the noise canceller, The control unit, Frequency analysis of the sound signal received through the plurality of microphones, and based on the preceding information to estimate whether each frequency response is the frequency response of the noise signal or the frequency response of the target signal, and using the estimated result Characterized by performing noise cancellation Noise Canceling Device. The method of claim 8, The noise removing unit A signal block forming unit dividing the sound signal into short-term multi-channel signal blocks in unit time units; A frequency division unit for performing frequency division filtering on the short-term multi-channel signal block in a time domain; A frequency response estimator for performing frequency analysis for each channel of the frequency-divided short-term multichannel signal block, and estimating a frequency response of the target signal based on direction information of each frequency; A noise processor for removing noise using a Wiener filter coefficient obtained in a time domain from the estimated frequency response of the target signal and the noise signal; A frequency synthesizer configured to perform frequency synthesis filtering on the target signal for each frequency band from which the noise is removed in the time domain; And And a signal end determination unit to determine whether the sound signal ends. Noise Canceling Device.

Images