KR101936242B1 - Apparatus and method for noise removal, and recording medium thereof - Google Patents

Abstract

Noise canceling apparatus and method, and recording medium are disclosed. The noise cancellation method according to an embodiment of the present invention includes: dividing a speech signal into a plurality of bands by wavelet packet decomposition; Calculating a standard deviation for each divided band and generating a threshold value matrix of bands based on the standard deviation calculated for each band; Determining a speech interval and a noise interval for each band based on at least one of a standard deviation and a threshold value matrix calculated for each band; Calculating a filtering mask having an index value as a standard deviation of a band determined as a noise interval for a band determined as a noise interval; And removing noise for a band determined as a noise period based on the filtering mask. According to the embodiment of the present invention, it is possible to improve voice continuity by effectively removing only noise while minimizing loss of voice information.

Description

TECHNICAL FIELD [0001] The present invention relates to a noise canceling apparatus, a noise canceling apparatus, a noise canceling apparatus,

The present invention relates to a noise cancellation apparatus and method, and a recording medium, and more particularly, to a noise cancellation apparatus, a noise cancellation method, and a recording medium that minimize noise while minimizing loss of audio information.

Recently, the degradation of system performance due to environmental noise has been recognized as an important problem to be solved in various voice signal processing applications, and a voice enhancement technology is required. The voice enhancement is a technique for improving the voice signal by removing noise from the voice signal and enhancing the voice when the voice signal is contaminated by the ambient noise, Or improve voice recognition or speaker recognition performance in human-machine interaction in medical devices such as smart devices or implantable hearing aids. It can also be used to suppress background noise and improve sound quality in audio equipment such as headsets and digital hearing aids.

To reduce the effect of noise, various noise attenuation techniques and speech enhancement techniques have been studied and used in various voice signal processing fields. For example, wavelet transform based algorithms are being studied for noise reduction and speech enhancement. However, in the conventional algorithm, when the noise characteristic is similar to the speaker's voice, or when the noise is distributed in a frequency band similar to or the same as the voice, it is impossible to divide the noise and the voice, There is a problem that it is difficult to secure.

Disclosed are a noise canceller, a method, and a recording medium that are excellent in noise cancellation performance and can effectively remove only noise while minimizing loss of voice information.

The problems to be solved by the present invention are not limited to the above-mentioned problems. Other technical subjects not mentioned will be apparent to those skilled in the art from the description below.

According to an aspect of the present invention, there is provided a noise removal method including: dividing a speech signal into a plurality of bands by wavelet packet decomposition; Calculating a standard deviation for each divided band and generating a threshold value matrix of the bands based on the standard deviation calculated for each band; Determining a speech interval and a noise interval for each band based on at least one of the standard deviation calculated for each band and the threshold value matrix; Calculating a filtering mask having a standard deviation of a band determined as the noise interval as an exponent value for a band determined as the noise period; And removing noise for a band determined as the noise period based on the filtering mask calculated for each band.

The step of generating the threshold value matrix may generate the threshold value matrix according to Equation (2) by calculating the standard deviation for each band in accordance with Equation (1) below.

[Equation 1]

[Equation 2]

는 상기 밴드들 중 m번째 밴드의 음성 신호값,

은 음성 신호의 샘플 길이,

은 상기 m번째 밴드의 표준편차, N은 음성 신호의 프레임당 샘플 개수,

은 상기 문턱값 매트릭스의 상기 m번째 밴드의 문턱값이다.In Equations (1) and (2)

Is the speech signal value of the m-th band among the bands,

Is the sample length of the speech signal,

Is the standard deviation of the m-th band, N is the number of samples per frame of the speech signal,

Is a threshold value of the m-th band of the threshold matrix.

Wherein the step of calculating the filtering mask comprises the step of calculating a filtering mask in which the standard deviation of the bands determined in the noise period and the standard deviation of the bands determined as the noise period are compared with the difference between the order of the fundamental bands, The filtering mask can be calculated to have a value proportional to the number of the filter coefficients.

Wherein the determining the speech interval and the noise interval comprises: comparing a magnitude of a speech signal of each band with a threshold value of the threshold matrix; Determining the voice interval as the voice interval if the voice signal size is equal to or greater than the threshold value; Comparing an average value of speech signals of a previous frame of the band with the standard deviation if the speech signal magnitude of the band is smaller than the threshold value; Determining a first noise interval if the average value of the speech signal of the previous frame is greater than the standard deviation; And determining a second noise interval when the average value of the speech signal of the previous frame is equal to or less than the standard deviation.

The removing the noise may include removing noise by applying the filtering mask to the band determined as the first noise period. And applying a voice signal of the band to a band determined to be the second noise period as '0'.

The step of removing the noise may remove the noise according to Equation (3).

[Equation 3]

는 상기 밴드들 중 m번째 밴드의 음성 신호,

은 상기 문턱값 매트릭스의 상기 m번째 밴드의 문턱값,

은 상기 m번째 밴드의 t-N 내지 t 구간의 N개의 음성 신호의 평균값, N은 음성 신호의 프레임당 샘플 개수,

은 상기 m번째 밴드의 표준편차, B는 상기 밴드들의 개수, I는 상기 기본 밴드의 순번,

는 상기 필터링 마스크이다.In Equation (3)

The audio signal of the m-th band among the bands,

Th band of the threshold value matrix,

Is an average value of N audio signals of the m-th band from tN to t, N is the number of samples per frame of the audio signal,

Is the standard deviation of the m-th band, B is the number of bands, I is the order of the base band,

Is the filtering mask.

According to another aspect of the present invention, there is provided a computer-readable recording medium on which a program for executing the noise reduction method is recorded.

According to another aspect of the present invention, there is provided a wavelet decomposition unit for decomposing a speech signal into a plurality of bands by wavelet packet decomposition. A threshold value calculation unit for calculating a standard deviation for each divided band and generating a threshold value matrix of the bands based on the standard deviation calculated for each band; A noise interval determining unit for determining a speech interval and a noise interval for each band based on at least one of the standard deviation calculated for each band and the threshold value matrix; A mask calculating unit for calculating a filtering mask having a standard deviation of a band determined by the noise period as an exponent value for a band determined as the noise period; And a noise removing unit for removing noise from a band determined as the noise period based on the filtering mask.

Wherein the mask calculator calculates a mask variation ratio of the bands determined by the noise interval and a ratio of the number of bands determined by the noise interval to the number of bands determined by the noise interval, The filtering mask may be calculated so that the value of the filtering mask is equal to the exponent value.

Wherein the noise interval determining unit compares a magnitude of a voice signal of each band with a threshold value of the threshold value matrix, and when the magnitude of the voice signal of the band is smaller than the threshold value, A second noise interval is determined when the average value of the speech signal of the previous frame is equal to or less than the standard deviation, have.

The noise canceller may remove the noise by applying the filtering mask to the bands determined to be the first noise period and may apply the audio signal of the band to zero for the band determined to be the second noise period have.

According to an embodiment of the present invention, there is provided a noise canceller, a method, and a recording medium that are excellent in noise cancellation performance and can effectively remove only noise while minimizing loss of voice information.

The effects of the present invention are not limited to the effects described above. Unless stated, the effects will be apparent to those skilled in the art from the description and the accompanying drawings.

1 is a flowchart of a noise removal method according to an embodiment of the present invention.
2 is a block diagram of a noise removing apparatus according to an embodiment of the present invention.
3 is a conceptual diagram showing a wavelet packet decomposition of a speech signal according to an embodiment of the present invention.
4 is a configuration diagram of a threshold value calculating unit constituting a noise removing apparatus according to an embodiment of the present invention.
FIG. 5 is an exemplary waveform of a white noise-free audio signal.
6 is an exemplary waveform of a white noise mixed speech signal (SNR: 5 dB).
FIG. 7 is a diagram illustrating a result of improving noise by eliminating noise according to an embodiment of the present invention. Referring to FIG.

Other advantages and features of the present invention and methods of achieving them will be apparent by referring to the embodiments described hereinafter in detail with reference to the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, and the present invention is only defined by the scope of the claims. Although not defined, all terms (including technical or scientific terms) used herein have the same meaning as commonly accepted by the generic art in the prior art to which this invention belongs. A general description of known configurations may be omitted so as not to obscure the gist of the present invention. In the drawings of the present invention, the same reference numerals are used as many as possible for the same or corresponding configurations. To facilitate understanding of the present invention, some configurations in the figures may be shown somewhat exaggerated or reduced.

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", "having", or "having" are intended to specify the presence of stated features, integers, steps, operations, components, Steps, operations, elements, parts, or combinations thereof, whether or not explicitly described or implied by the accompanying claims.

Used throughout this specification may refer to a hardware component such as, for example, software, FPGA or ASIC, as a unit for processing at least one function or operation. However, "to" is not meant to be limited to software or hardware. &Quot; to " may be configured to reside on an addressable storage medium and may be configured to play one or more processors.

As an example, the term '~' includes components such as software components, object-oriented software components, class components and task components, and processes, functions, attributes, procedures, Routines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables. The functions provided by the components and components may be performed separately by a plurality of components and components, or may be integrated with other additional components.

The noise cancellation method according to an embodiment of the present invention includes: dividing a speech signal into a plurality of bands by wavelet packet decomposition; Calculating a standard deviation for each divided band and generating a threshold value matrix of bands based on the standard deviation calculated for each band; Determining a speech interval and a noise interval for each band based on at least one of a standard deviation and a threshold value matrix calculated for each band; Calculating a filtering mask having an index value as a standard deviation of a band determined as a noise interval for a band determined as a noise interval; And removing a noise for a band determined as a noise period based on the filtering mask calculated for each band.

1 is a flowchart of a noise removal method according to an embodiment of the present invention. 2 is a block diagram of a noise removing apparatus according to an embodiment of the present invention. 1 and 2, the voice signal input unit 110 receives a voice signal including noise (S10). In one embodiment, the voice signal input unit 110 may be provided as a device for receiving a voice signal, such as a microphone. As another example, the voice signal input unit 110 may be provided as a communication interface device for receiving a voice signal from another device.

When a voice signal mixed with noise is input from the voice signal input unit 110, the wavelet packet decomposition unit 120 decomposes the voice signal into bands of a time-frequency two-dimensional region by wavelet packet decomposition (S20).

According to one embodiment, the wavelet packet decomposition unit 120 may convert the speech signal into a speech signal according to the magnitude of the energy stimulated in the human auditory nerve based on the human acoustic acoustic model, And can be decomposed into 20 bands as shown in Fig.

The bands decomposed from the speech signal have the form of a time domain signal having 20 frequency information, and both time and frequency information can be represented as a two dimensional matrix.

In one embodiment, the speech signal may be decomposed into wavelet coefficients w _{j, m} ( k ) having 20 subbands. The wavelet coefficients w _{j , m} ( i ) represent the i- th wavelet coefficients of the j- th level and the m- th wavelet band ( j = 3, 4, 5, m = 1,. In order to process the wavelet coefficients w _{j, m} ( i ) simultaneously in the time and frequency domain, the wavelet coefficients can be represented by a two-dimensional matrix as shown in the following equation (1).

[Formula 1]

는 특정시간 t에서의 m번째 서브밴드의 웨이블릿 계수를 나타낸다. 다시 도 1 및 도 2를 참조하면, 웨이블릿 패킷 분해부(120)에 의해 음성 신호가 다수의 밴드들로 분해되면, 문턱값 산출부(130)는 분할된 밴드별로 표준편차를 산출하고, 밴드별로 산출된 표준편차를 기반으로 밴드들의 문턱값 매트릭스를 생성한다.here

Represents the wavelet coefficient of the mth subband at a specific time t . 1 and 2, when the speech signal is decomposed into a plurality of bands by the wavelet packet decomposition unit 120, the threshold value calculation unit 130 calculates a standard deviation for each divided band, And generates a threshold matrix of bands based on the calculated standard deviation.

4 is a configuration diagram of a threshold value calculating unit constituting a noise removing apparatus according to an embodiment of the present invention. 1, 2, and 4, the threshold calculator 130 calculates a standard deviation for each divided band, and calculates a threshold matrix of bands based on the standard deviation calculated for each band (S30).

In one embodiment, the threshold calculator 130 may include a standard deviation calculator 132 and a threshold matrix generator 134. [ In one embodiment, the standard deviation calculator 132 may calculate the standard deviation for each band according to the following equation (2).

[Formula 2]

는 상기 밴드들 중 m번째 밴드의 음성 신호값,

은 음성 신호의 샘플 길이(예를 들어, 25ms),

은 상기 m번째 밴드의 표준편차이다.In Equation 2,

Is the speech signal value of the m-th band among the bands,

(E.g., 25 ms) of the speech signal,

Is the standard deviation of the m-th band.

When the standard deviation is calculated for each band, the threshold matrix generator 134 can generate a threshold matrix according to Equation 3 below.

[Formula 3]

(m=1,2,...,20)

(m = 1, 2, ..., 20)

은 m번째 밴드의 표준편차, N은 음성 신호의 프레임당 샘플 개수,

은 문턱값 매트릭스(threshold matrix)의 m번째 밴드의 문턱값이다.In Equation 3,

Is the standard deviation of the m-th band, N is the number of samples per frame of the speech signal,

Is the threshold of the mth band of the threshold matrix.

Referring again to FIGS. 1 and 2, the noise interval determining unit 140 determines a voice interval and a noise interval for each band based on the standard deviation and the threshold value matrix of each band calculated by the threshold calculating unit 130 (S40).

In one embodiment, the noise interval determining unit 140 compares the voice signal size of each band with the threshold value of the threshold value matrix, and determines that the voice interval is equal to or larger than the threshold value.

The noise interval determining unit 140 compares the average value of the voice signals of the previous frame of the band with the standard deviation when the voice signal size of the band is smaller than the threshold, 1 noise period, and when the average value of the speech signal of the previous frame is equal to or less than the standard deviation, it is determined to be the second noise period.

The mask calculating unit 150 calculates a filtering mask having the standard deviation of the band determined by the first noise period as an exponent value for the band determined as the first noise period by the noise period determining unit 140 (S50).

In an exemplary embodiment, the mask calculator 150 may be configured to calculate a mask band inversely proportional to the difference between the order of the fundamental wavelet band, which is the highest energy band among the bands, and the order of bands determined by the noise period, The filtering mask can be calculated so as to have a value proportional to the standard deviation and the number of bands as an exponent value.

The mask calculating unit 150 may calculate energy for each band for determining a base band, and determine a band having the largest energy among the bands as a base band. The energy of the band can be calculated based on the average signal size, entropy, standard deviation or dispersion of the band, but is not limited thereto.

The noise removing unit 160 removes noise for a band determined as a noise period based on the filtering mask calculated by the mask calculating unit 150 (S60). In one embodiment, the noise removing unit 160 removes noise by applying a filtering mask to bands determined to be the first noise period by the noise period determining unit 140, The audio signal of the band can be applied as zero. In one embodiment, the noise eliminator 160 may remove noise for each band according to Equation (4) below.

[Formula 4]

는 m(m=1,2,...,B)번째 밴드의 음성 신호,

은 문턱값 매트릭스의 m번째 밴드의 문턱값,

은 m번째 밴드의 t-N 내지 t 구간의 N개의 음성 신호의 평균값, N은 음성 신호의 프레임당 샘플 개수,

은 m번째 밴드의 표준편차, B는 밴드들의 개수(도 3의 예에서 20개), I는 밴드들 중 가장 에너지가 큰 밴드인 기본 밴드의 순번,

는 필터링 마스크이다. 밴드의 순번은 저주파수 대역 순으로 1,2,...,B로 순차적으로 할당될 수 있다.In Equation (4)

(M = 1, 2, ..., B) -th band audio signal,

Is the threshold value of the m < th > band of the threshold value matrix,

Is an average value of N audio signals in the interval of tN to t of the m-th band, N is the number of samples per frame of the audio signal,

B is the number of bands (20 in the example of FIG. 3), I is the order of the base band which is the band with the highest energy among the bands,

Is a filtering mask. The order of the bands can be sequentially assigned to 1, 2, ..., B in the order of the low frequency bands.

가 적용되어 음성 손실을 최소화하면서 잡음이 제거된다. 세미소프트 필터링 마스크는 각 밴드별 특성을 반영하여 잡음을 제거하도록 산출된다.According to Equation (4), no noise cancellation is performed for a speech interval (a speech signal having a threshold value or more) for each band, a speech signal value is applied to a second noise interval, For the first noise interval between the intervals, a semi-soft filtering mask (exponential filtering)

Is applied to eliminate noise while minimizing voice loss. The semi-soft filtering mask is calculated so as to remove noise by reflecting characteristics of each band.

According to the present embodiment, as the band of the frequency band close to the frequency of the band having the highest energy (fundamental band), that is, the absolute value of (Im) in Equation 4 is smaller, the filtering mask value exponentially increases, Elimination performance is reduced, so that speech preservation performance can be improved if noise and speech are distributed within similar bands. On the other hand, in the case of a band having a frequency band far from the fundamental band, the absolute value of (Im) in Equation (4) increases and the filtering mask value exponentially decreases, Effectively removed.

In the case of dividing a signal mixed with speech and noise using the wavelet decomposition according to the present embodiment, since the intensity of the entire signal is stronger than the noise in the divided bands, the coefficient value of the wavelet band is larger than noise . As a result, the standard deviation and the threshold value (Π _m ) of the corresponding band are large in the band of the voice band in which the signal strength is large. In this case, the semi-soft filtering mask is calculated to have a relatively large value, thereby reducing loss of voice information. In the opposite case, the semi-soft filtering mask is calculated to have a relatively small value, thereby improving the noise canceling performance.

As described above, according to the present embodiment, the standard deviation of each band is calculated by using the speech region information of the wavelet packet decomposed bands, and the threshold value (Π _m ) for distinguishing speech and noise is calculated It is decided flexibly for each band. In addition, according to a threshold value that is fluidly determined for each band, a criterion of a speech interval, a first noise interval (an intermediate interval to which a semi-soft filtering mask is applied) and a second noise interval are changed, Instead of applying a zero value to the speech signal, the region of speech and noise may be set to be gentle, and the noise in the intermediate region (first noise region) may be effectively removed by the filtering mask. Accordingly, it is possible to effectively remove noise only by minimizing loss of voice information in a noise process in a region where noise and voice are similarly distributed (first noise interval), and improve voice continuity.

In order to verify the performance of the noise canceller according to the embodiment of the present invention, noise cancellation experiments were performed under various noise environments using signal samples arbitrarily extracted from an authorized database. Samples of the speech signal were extracted from the TIMIT database and samples of the noise signal were extracted from the NOISEX-9213 database. The extracted data samples have a sampling rate of 16 kHz and a bit rate of 16 bps. Experiments were carried out under various SNR conditions (0dB, 5dB, 10dB, 15dB) by mixing various noises such as white noise, pink noise, and tank engine noise.

6 is an exemplary waveform of a white noise mixed speech signal (SNR: 5 dB), and FIG. 7 is a flowchart illustrating a method of removing noise according to an embodiment of the present invention Fig. 8 is a diagram showing a result of improving speech. As shown in Fig. 7, according to the present embodiment, it is confirmed that the noise is removed cleanly, and the sound is improved so as to exhibit substantially the same waveform as in Fig.

As shown in Table 1, according to the embodiment of the present invention, excellent noise cancellation and speech enhancement performance are exhibited in various noise environments. Particularly, the embodiment of the present invention shows good results even in a SNR environment with a lot of noises. The speech enhancement technology according to an embodiment of the present invention can be applied to various voice signal processing fields such as an implantable digital hearing aid, voice communication, human-machine interaction, and smart device.

The method according to an embodiment of the present invention can be realized in a general-purpose digital computer that can be created as a program that can be executed by a computer and operates the program using a computer-readable recording medium. The computer readable recording medium may be a volatile memory such as SRAM (Static RAM), DRAM (Dynamic RAM), SDRAM (Synchronous DRAM), ROM (Read Only Memory), PROM (Programmable ROM), EPROM (Electrically Programmable ROM) Non-volatile memory such as EEPROM (Electrically Erasable and Programmable ROM), flash memory device, Phase-change RAM (PRAM), Magnetic RAM (MRAM), Resistive RAM (RRAM), Ferroelectric RAM But are not limited to, optical storage media such as CD ROMs, DVDs, and the like.

It is to be understood that the above-described embodiments are provided to facilitate understanding of the present invention, and do not limit the scope of the present invention, and it is to be understood that various modified embodiments are also within the scope of the present invention. It is to be understood that the technical scope of the present invention should be determined by the technical idea of the claims and the technical scope of protection of the present invention is not limited to the literary description of the claims, To the invention of the invention.

100: Noise canceling device
110: audio signal input unit
120: wavelet packet decomposition unit
130: Threshold value calculating section
140: noise section determining section
150: mask calculation unit
160: Noise canceling

Claims (11)

Dividing a speech signal into a plurality of bands by wavelet packet decomposition;
Calculating a standard deviation for each divided band and generating a threshold value matrix of the bands based on the standard deviation calculated for each band;
Determining a speech interval and a noise interval for each band based on at least one of the standard deviation calculated for each band and the threshold value matrix;
Calculating a filtering mask having a standard deviation of a band determined as the noise interval as an exponent value for a band determined as the noise period; And
And removing noise for a band determined as the noise period based on the filtering mask,
Wherein the step of calculating the filtering mask comprises:
Wherein the filtering mask is calculated based on the order of the bands determined based on the frequency band of the bands and the standard deviation of bands determined by the noise period. Dividing a speech signal into a plurality of bands by wavelet packet decomposition;
Calculating a standard deviation for each divided band and generating a threshold value matrix of the bands based on the standard deviation calculated for each band;
Determining a speech interval and a noise interval for each band based on at least one of the standard deviation calculated for each band and the threshold value matrix;
Calculating a filtering mask having a standard deviation of a band determined as the noise interval as an exponent value for a band determined as the noise period; And
And removing noise for a band determined as the noise period based on the filtering mask,
The step of generating the threshold value matrix may include: calculating a standard deviation for each band according to Equation 1 below, generating the threshold value matrix according to Equation 2 below,
[Equation 1]

[Equation 2]

In Equations (1) and (2)

Is the speech signal value of the m-th band among the bands,

Is the sample length of the speech signal,

Is the standard deviation of the m-th band, N is the number of samples per frame of the speech signal,

Is a threshold value of the m-th band of the threshold matrix. The method according to claim 1,
Wherein the step of calculating the filtering mask comprises:
A value that is inversely proportional to a difference between a sequence number of a base band that is the highest energy band among the bands and a sequence number of bands determined as the noise period and that is proportional to a standard deviation of the band determined by the noise interval, Value of the filtering mask. The method of claim 3,
Wherein the determining the speech interval and the noise interval comprises:
Comparing a voice signal size of each band with a threshold value of the threshold value matrix;
Determining the voice interval as the voice interval if the voice signal size is equal to or greater than the threshold value;
Comparing an average value of speech signals of a previous frame of the band with the standard deviation if the speech signal magnitude of the band is smaller than the threshold value;
Determining a first noise interval if the average value of the speech signal of the previous frame is greater than the standard deviation; And
And determining a second noise period if the average value of the speech signals of the previous frame is equal to or less than the standard deviation. 5. The method of claim 4,
The step of removing the noise includes:
Removing noise by applying the filtering mask to bands determined as the first noise period; And
And applying a voice signal of the band to zero for a band determined as the second noise period. 5. The method of claim 4,
The noise removing step may include removing noise according to Equation (3) below,
[Equation 3]

In Equation (3)

The audio signal of the m-th band among the bands,

Th band of the threshold value matrix,

Is an average value of N audio signals of the m-th band from tN to t, N is the number of samples per frame of the audio signal,

Is the standard deviation of the m-th band, B is the number of bands, I is the order of the base band,

Wherein the filtering mask is the filtering mask. A computer-readable recording medium having recorded thereon a program for executing the noise cancellation method according to any one of claims 1 to 6. A wavelet packet decomposition unit for dividing a speech signal into a plurality of bands by wavelet packet decomposition;
A threshold value calculation unit for calculating a standard deviation for each divided band and generating a threshold value matrix of the bands based on the standard deviation calculated for each band;
A noise interval determining unit for determining a speech interval and a noise interval for each band based on at least one of the standard deviation calculated for each band and the threshold value matrix;
A mask calculating unit for calculating a filtering mask having a standard deviation of a band determined by the noise period as an exponent value for a band determined as the noise period; And
And a noise eliminator for removing noise with respect to a band determined as the noise period based on the filtering mask,
The mask calculating unit may calculate,
Wherein the filtering mask is calculated based on the order of the bands determined based on the frequency band of the bands and the standard deviation of the band determined by the noise period. 9. The method of claim 8,
The mask calculating unit may calculate,
A value that is inversely proportional to a difference between a sequence number of a base band that is the highest energy band among the bands and a sequence number of bands determined as the noise period and that is proportional to a standard deviation of the band determined by the noise interval, Value of the filtering mask. 10. The method of claim 9,
The noise interval determining unit may determine,
Comparing the average value of the speech signal of the previous frame of the band with the standard deviation when the size of the speech signal of the band is smaller than the threshold value, And determines the second noise interval as a first noise interval when the average value of the speech signal of the previous frame is greater than the standard deviation and determines the second noise interval as the average value of the speech signal of the previous frame is equal to or less than the standard deviation. 11. The method of claim 10,
The noise-
Wherein the filtering mask is applied to the band determined as the first noise period to remove noise and the speech signal of the band is applied to the band determined to be the second noise period as zero.

Images