KR100744375B1 - Apparatus and method for processing sound signal - Google Patents

Abstract

The speech processing apparatus according to the present invention comprises a speech signal input unit for receiving a speech signal, a frequency domain converter for converting the input speech signal into the frequency domain, and a harmonic part and a noise part separated from the speech signal converted into the frequency domain. Signal processed by harmonic-noise separation section, noise suppression index determination section that determines the optimum noise suppression index k according to the situation and system, and suppression of the noise parts separated in the harmonic-noise separation section according to the noise suppression index It includes a noise suppressor for outputting.

Harmonic, noise

Description

Speech processing device and method {APPARATUS AND METHOD FOR PROCESSING SOUND SIGNAL}

1 is a block diagram illustrating a speech processing apparatus according to an embodiment of the present invention;

2 is a diagram illustrating a speech signal on a frequency domain;

3 is a view for explaining a voice processing method according to an embodiment of the present invention;

4 is a diagram illustrating an internal configuration diagram of a harmonic-noise separation unit in a speech processing device according to an embodiment of the present invention;

5 is a view for explaining a harmonic-noise separation method according to an embodiment of the present invention;

6 is a diagram illustrating signals of a harmonic section and a noise section separated according to an embodiment of the present invention.

The present invention relates to a speech processing device and a method.

In all speech signal processing applications, noise processing is the most important and difficult problem.

The conventional noise processing algorithm expects the noise removal effect after applying the algorithm in a predetermined method according to a specific algorithm, and does not consider the flexibility and utilization according to various noises and situations. Most of the existing noise processing methods are different from the application method, and apply a kind of filtering concept. In addition, conventional noise reduction methods can handle noise based on various assumptions, and often fail in situations that do not fit the assumptions. Therefore, there are not so many commercially available noise reduction and control algorithms that can be applied to too many and various real world noises.

Accordingly, the present invention provides a speech processing apparatus and method capable of efficiently removing noise in various cases.

The present invention also provides a speech processing apparatus and method for accurately distinguishing between harmonic and non-harmonic sections of a speech signal.

To this end, the voice processing apparatus according to the present invention includes a voice signal input unit for receiving a voice signal, a frequency domain converter for converting the input voice signal into a frequency domain, and a harmonic part and noise from the voice signal converted into the frequency domain. A harmonic-noise separation unit for separating the portions, a noise suppression index determination unit for determining an optimal noise suppression index k according to a situation and a system, and a noise portion separated from the harmonic-noise separation unit according to the noise suppression index. And a noise suppression unit for outputting a noise processed signal by suppressing the noise.

In addition, the voice processing method according to the present invention comprises the steps of converting the input voice signal into the frequency domain when the voice signal is input, separating the harmonic part and the noise part from the voice signal converted into the frequency domain, the situation And determining an optimal noise suppression index k according to the system, and suppressing the separated noise portion according to the noise suppression index and outputting a noise processed signal.

In addition, in the apparatus for processing a speech signal according to the present invention, the harmonic is input until the energy difference between the speech signal input unit for receiving the speech signal and the energy difference of two consecutive harmonic components from the input speech signal is less than or equal to a preset threshold value. A harmonic-noise separation section for separating the harmonic portion and the noise portion when the energy difference between the two consecutive harmonic components falls below a predetermined threshold value by repeatedly performing amplification of the portion and attenuation of the noise portion. The noise suppression index determination unit for determining an optimum noise suppression index k according to the noise suppression index, and the noise suppression unit for outputting a noise processed signal by suppressing the separated noise portion in accordance with the noise suppression index.

In addition, in the speech processing method of the present invention, a process of repeatedly performing amplification and attenuation of a noise part of a harmonic part until a time point at which an energy difference between two consecutive harmonic components is equal to or less than a preset threshold value from an input voice signal And separating the harmonic part and the noise part when the energy difference between two consecutive harmonic components after the amplification of the harmonic part and the attenuation of the noise part is less than or equal to a preset threshold value, according to the situation and system. Determining an optimal noise suppression index k, and suppressing the separated noise portion according to the noise suppression index and outputting a noise processed signal.

The present invention is applicable to a system in which any speech signal processing including speech coding, synthesis, and recognition is applied, and optimal noise processing is flexibly adapted to the system as needed without any assumptions about situations, signals, and types of noise. Presents a possible new algorithm.

The present invention proposes a method for separating harmonics and noise sections and suppressing noise using optimal parameters for noise suppression for the noise sections. At this time, the optimal variable for noise suppression may be set in advance according to the system or automatically set by the system in accordance with the situation. In real life voice signals are mostly varied and unpredictable noise, and all sounds other than the intended speaker's voice are considered noise. However, the unilateral processing method of the existing noise processing algorithm fails in many cases when the situation is not suitable for the algorithm. In addition, there is no noise processing algorithm that the user can arbitrarily determine the desired level of noise processing for the system, and moreover, the noise processing factor can be adaptively optimized and flexibly used. There is no algorithm that can be a source technology. Therefore, in the present invention, a flexible, adaptive method that can be widely and easily applied to any voice signal related system, a simple, yet robust, creative approach using the optimal parameters desired by the user based on a new theoretical basis. We present a speech noise processing technique.

Then, with reference to the accompanying drawings, preferred embodiments of the present invention will be described in detail. In addition, detailed descriptions of well-known functions and configurations that may unnecessarily obscure the subject matter of the present invention will be omitted.

The present invention discloses a configuration in which a speech signal is divided into a harmonic section and a noise section and the noise section is suppressed according to a noise suppression index suitable for a system or a situation.

1 is a block diagram illustrating a speech processing apparatus according to an embodiment of the present invention. Referring to FIG. 1, an apparatus for processing a speech signal according to an exemplary embodiment of the present invention may include a speech signal input unit 110, a frequency domain converter 120, a harmonic-noise separator 130, a noise suppressor 140, and an optimum. Noise suppression index determination unit 150 of FIG.

The voice signal input unit 110 may be configured as a microphone (MIC) and receives a voice signal. The frequency domain converter 120 converts the input voice signal from the time domain to the frequency domain. The frequency domain transformer 120 converts a voice signal in the time domain into a voice signal in the frequency domain by using a fast fourier transform (FFT).

The harmonic-noise separator 130 separates the speech signal on the frequency domain into a harmonic section and a noise section. The harmonic-noise separation unit 130 selects a predetermined length sample frame of a linear prediction residual signal from the frequency domain converter 120 and converts the input speech signal into a frequency range of a predetermined section. Is input. Referring to FIG. 4, the operation according to the detailed configuration and configuration will be described with reference to FIG. 4 for the harmonic-noise separation unit 130 for separating the harmonic section and the noise section according to the present invention. The harmonic-noise separation unit 130 according to the present invention includes a harmonic section determination unit 400, a harmonic extrapolation unit 401, a noise estimation unit 402, and a noise extrapolation unit. 404, a harmonic-noise separation iteration 407 including a harmonic estimation unit 406, and a harmonic-noise section separated from the harmonic-noise separation determination unit 408 A harmonic-noise interval extractor 409 may be included.

First, the harmonic section determination unit 400 determines a harmonic section using cepstrum and pitch information when a voice signal converted into the frequency domain is input from the frequency domain converter 120.

Next, the voice signal on the frequency domain will be described with reference to FIG. 2. 2 is a diagram illustrating a voice signal on a frequency domain. Referring to FIG. 2, the speech signal may be separated into a noise section B 10 and a harmonic section A 20. Conventionally, since the noise is filtered from the speech signal according to the noise level in the speech signal, the harmonic section A 20 is also suppressed, thereby affecting the quality of the speech signal. However, in the present invention, the noise is suppressed only in the noise section except the harmonic section.

Herein, if the voice signal is x (n), the harmonic part is h (n), and the noise part is w (n), the voice signal may be expressed as in Equation 1 below.

On the other hand, the harmonic-noise separation repeater 407 performs interpolation and extrapolation between the harmonic section and the noise section. The harmonic section and the noise section are repeated until the harmonic section and the noise section are correctly separated. Perform interpolation and extrapolation. The harmonic-noise separation repeater 407 includes a harmonic extrapolation 401, a noise estimator 402, a noise extrapolation 404, and a harmonic estimator. (406).

The harmonic extrapolator 401 sets frequency domain expression values (eg, DFT values) of noise sections other than the harmonic section determined by the harmonic section determination unit 400 to '0'.

The noise determiner 402 extrapolates the current harmonic or sinusoidal samples of the harmonic or sinusoidal regions to the noise region. Subtract the harmonic samples of the noise intervals from the initial noise sample, and extrapolate the residual noise sample estimates into the harmonic or sinusoidal region (harmonic (or sinusoidal) region).

In this case, the 'initial noise sample' refers to the original linear prediction residual spectrum of the noise region.

On the other hand, the noise extrapolation unit 404 sets the DFT value to 0, for example, frequency domain representation values of the harmonic region.

The harmonic determination unit 406 extrapolates the current noise samples of the noise region to the harmonic region. (extrapolate) The noise samples of the harmonic intervals are subtracted from the initial harmonic sample that has undergone the harmonic interval extrapolation, and the residual harmonic sample estimates are extrapolated to the noise interval. (extrapolate)

At this time, the 'initial harmonic sample' refers to the original linear prediction residual spectrum of the harmonic region.

As described above, the harmonic signal of the harmonic section is amplified in the frequency domain through the harmonic-noise separation repeater 407 and attenuates the noise signal of the noise section.

Subsequently, when the harmonic signal of the harmonic section is amplified in the frequency domain of the input voice signal as described above and the noise signal of the noise section is attenuated, the harmonic-noise separation determination unit 408 performs two consecutive harmonic components. It is determined whether the energy difference of these drops below a preset threshold. Also, the harmonic-noise separation determination unit 408 may use the harmonic extrapolation unit 401 and the noise determination unit 402 until the determination result of the energy difference between two consecutive harmonic components becomes less than or equal to a threshold. By repeating the noise extrapolation unit 404 and the harmonic determination unit 406, the harmonic section is amplified and the noise section is attenuated. In addition, the harmonic-noise separation determination unit 408 classifies the harmonic noise section extraction unit 409 according to the amplification and attenuation when the energy difference between two consecutive harmonic components is equal to or less than a threshold. The harmonic section and the noise section are separated, and each harmonic noise section is provided to a next noise suppressor 140.

The speech signal of the harmonic section and the noise section separated by the harmonic-noise section extractor 409 may be illustrated in FIG. 6 in the frequency domain. Referring to FIG. 6, (a) is a harmonic signal including a harmonic period, and (b) is a non-harmonic component including a noise period. Through the harmonic-noise separation unit 130 of the present invention, it can be seen that as shown in Figure 6 (a), (b) can be separated accurately. As described above, the method of separating the harmonic part and the noise part in the frequency band may be variously used in a system using all voice and audio signals such as coding, synthesis, recognition, and enhancement.

When the harmonic-noise section is separated by the harmonic-noise separation unit 130 as described above, the noise suppression unit 140 uses a system for implementing a speech processing device or a noise suppression index according to the characteristics of the noise region. Repress

라 할 때 잡음 처리된 신호 는 다음 수학식 2와 같이 표시할 수 있다.As described above, the noise suppression unit 140 generates a noise reduced signal using an optimal suppression index for the noise section.

In this case, the noise processed signal may be expressed as Equation 2 below.

Here, k is a noise processed signal, k is an optimum noise suppression index for optimally suppressing noise according to the system in which the speech processing apparatus is implemented and its characteristics, h is a harmonic part, and w is a noise part.

The optimum noise suppression index determination unit 150 determines the noise suppression index. The noise suppression index indicates how much noise is to be suppressed. Noise is undesirably suppressed because the frequency domain non-harmonic component also contains components of the speech signal. Accordingly, the present invention determines the noise suppression index according to the system in which the speech processing device is implemented or its characteristics.

를 구한다. 이 경우 본 발명은 아래의 두가지 필수적인 법칙(constraint)을 적용한다.That is, the present invention determines the k (degree of noise reduction as desired in the system) from the original signal x (n) and then the signal after noise processing.

Obtain In this case, the present invention applies the following two essential constraints.

1. The signal before and after noise processing has the same energy.

2. The signal after noise processing is close to the signal before noise processing.

이고, 두번째 법칙은

이다.(여기서β<1, k<1)The first law is

And the second law is

(Where β <1, k <1)

If the above rules are applied to the speech signal of each frame in vector form, the following equation (3) is obtained.

,

,

Therefore, the following equation (4) is arranged.

도 자연스럽게 구해진다. 이러한 본 발명은 음성 신호에서 하모닉 부분과 잡음 부분을 분리한 후에 아주 쉽게 적용 가능하며 당업자에 의해 유연하게 적용될 수 있다. 즉, 본 발명은 상황과 시스템에 따라 최적의 잡음 억압 지수 k를 임의로 사용할 수 있으므로 적응적으로 적용 가능하다.Thus, according to the present invention, k (<1) is input according to the desired noise suppression degree, and K accordingly is obtained automatically. Therefore signal after noise processing

Also naturally obtained. This invention is very easily applicable after separating the harmonic part and the noise part from the speech signal and can be flexibly applied by those skilled in the art. That is, the present invention can be adaptively applied because the optimum noise suppression index k can be arbitrarily used according to the situation and the system.

는 다음 수학식 5와 같이 구해질 수 있다. Thus, with K

May be obtained as in Equation 5 below.

According to the noise suppression index obtained as described above, the noise suppression unit 140 suppresses and outputs the noise section B 10 of the speech signal. At this time, since the harmonic region and the noise region are processed separately by the harmonic-noise separation unit 130, the harmonic region and the noise region are processed separately, so that the speech signal in the state of suppressing the noise is the harmonic region and the suppressed noise. Outputs the signal including the area.

Next, a voice processing method according to an embodiment of the present invention will be described with reference to FIG. 3.

Referring to FIG. 3, the voice signal input unit 110 of the voice processing apparatus 100 receives a voice signal through a microphone in step 210. In operation 220, the frequency domain transform unit 120 converts the input voice signal into the frequency domain using the fast fourier transform (FFT). The harmonic-noise separation unit 130 then separates the harmonic and noise portions from the speech signal on the frequency domain in step 230. An operation of separating the harmonic part and the noise part from the voice signal in step 230 will be described in more detail with reference to FIG. 5 below. Thereafter, the speech processing apparatus 100 determines the optimum noise suppression index in operation 240 through the optimal noise suppression index determination unit 150. As mentioned above, the noise suppression index indicates how much noise is to be suppressed. Noise is undesirably suppressed because the frequency domain non-harmonic component also contains components of the speech signal. Accordingly, the present invention determines the noise suppression index according to the system in which the speech processing device is implemented or its characteristics.

Subsequently, the speech processing apparatus 100 may obtain the speech signal from which the noise is removed by noise suppressing the noise portion of the speech signal according to the optimal noise suppression index obtained in operation 250 in operation 250.

Then, the process in which the harmonic-noise separation unit 130 separates the harmonic section and the noise section from the speech signal in step 230 of FIG. 3 will be described in more detail with reference to FIG. 5. 5 illustrates a harmonic-noise separation method according to an embodiment of the present invention.

In operation 500, when the voice signal converted into the frequency domain is input from the frequency domain converter 120, the harmonic section determination unit 400 determines a harmonic section using cepstrum and pitch information. .

In operation 502, the harmonic extrapolation unit 401 sets the frequency domain expression values of noise sections other than the harmonic section determined by the harmonic section determination unit 400 to '0'.

In operation 502, the noise determination unit 402 may generate current harmonic or sinusoidal samples of the harmonic or sinusoidal regions. Extrapolate with (extrapolate)

In operation 506, the noise determiner 402 subtracts the harmonic samples of the noise intervals from the initial noise sample that has undergone the noise interval estimation, and performs a harmonic or a residual noise sample estimate. Extrapolate to the sine wave region (harmonic or sinusoidal region) (extrapolate).

In this case, the 'initial noise sample' refers to the original linear prediction residual spectrum of the noise region.

That is, the voice processing apparatus 100 performs an operation of amplifying the voice signal of the harmonic section through steps 502 to 506.

In operation 508, the noise extrapolator 404 sets a frequency domain representation value of the harmonic region determined by the harmonic interval determination unit 400, for example, sets a DFT value to '0', and then, in operation 510. The harmonic determination unit 406 extrapolates the current noise samples of the noise region to the harmonic region. After extrapolate, the harmonic determination unit 406 subtracts the noise samples of the harmonic intervals from the initial harmonic sample in step 512, and extrapolates the remaining harmonic sample estimates to the noise interval. Do the law. (extrapolate). In this case, the 'initial harmonic sample' refers to the original linear prediction residual spectrum of each harmonic region.

That is, the voice processing apparatus 100 performs an operation of attenuating the voice signal in the noise section through steps 508 to 512.

Thereafter, the speech processing apparatus 100 amplifies the speech signal in the harmonic section of the speech signal input through the above steps 502 to 512, attenuates the speech signal in the noise section, and proceeds to step 514.

In operation 514, the harmonic-noise separation determination unit 408 determines whether an energy difference between two consecutive harmonic components falls below a preset threshold.

If the energy difference between the two consecutive harmonic components is less than or equal to the threshold as a result of the determination in step 514, the process proceeds to step 516. The harmonic noise section extractor 409 divides the harmonic section and the noise according to the amplification and attenuation. The sections are separated and each harmonic noise section is provided to the next stage, the noise suppressor 140.

However, when the determination result of step 514 indicates that the energy difference between two consecutive harmonic components exceeds a threshold, the flow proceeds to step 502 and the steps 502 to 512 are continued until the energy difference between two consecutive harmonic components falls below a threshold. By repeating, the harmonic section is amplified and the noise section is attenuated.

The algorithm proposed in the present invention is applicable to all speech signal processing systems and may be a source technology of speech signal processing that performs a speech enhancement function.

를 각 음성 신호 처리 시스템이 입력 신호로 사용할 수 있다. 특히, 휴대 단말기(mobile phone)과 텔레메틱스(telematics) 등의 시스템과 같이 이동에 따른 다양한 상황의 잡음이 문제가 될 때, 종래의 잡음 처리 방식으로는 예측 불가능한 많은 상황에 맞는 최적의 잡음 처리가 불가능했으나, 본 발명의 방법은 원하는 잡음 처리 정도를 시스템이 결정할 수 있도록 함으로써 이러한 문제점을 해결하였다. 뿐만 아니라, 이 기술은 음성 신호 처리 시스템의 중간에도 쉽게 삽입하여 시스템 효율은 높일 수가 있으며, 음성 처리 시스템의 후처리부(Back-End or Post-processing)에 삽입하면 잡음 제거를 통해 음질을 향상 시킬 수 있다. 이와 같이 알고리듬 자체만이 아니라, 그 적용에 있어서도 매우 유동적(flexible)이며 활용성이 높은 기술이다.For example, in voice coding, synthesis, and recognition algorithms, it can be easily inserted into the front-end or pre-processing system and the desired noise suppression index, k, depending on the requirements and specifications of the back-end system. Or adaptively inputs the optimum noise suppression index k value, thereby reducing the noise reduced signal.

Can be used as an input signal by each voice signal processing system. In particular, when noise of various situations due to movement becomes a problem, such as systems such as mobile phones and telematics, it is impossible to optimize noise for many situations that are unpredictable by conventional noise processing methods. However, the method of the present invention solves this problem by allowing the system to determine the desired degree of noise processing. In addition, this technology can be easily inserted in the middle of a voice signal processing system to increase system efficiency. When inserted into the back-end or post-processing part of a voice processing system, the sound quality can be improved by removing noise. have. Thus, not only the algorithm itself, but also its application is very flexible and highly available.

In the above description of the present invention, specific embodiments have been described, but various modifications may be made without departing from the scope of the present invention. Therefore, the scope of the present invention should not be defined by the described embodiments, but should be determined by the equivalent of claims and claims.

The present invention determines the level of noise reduction as much as the user desires for noise processing, which is the most important and difficult problem in all speech signal processing related systems including speech recognition. Can be implemented.

Claims (19)

In the apparatus for processing a voice signal, A voice signal input unit for receiving a voice signal, A frequency domain converter for converting the input voice signal into a frequency domain; A harmonic-noise separator for separating the harmonic part and the noise part from the speech signal converted into the frequency domain; A noise suppression index determination unit for determining an optimal noise suppression index k according to a situation and a system; And a noise suppressor for outputting a noise processed signal by suppressing a noise part separated by the harmonic-noise separator according to the noise suppression index. The speech processing apparatus of claim 1, wherein the harmonic-noise separation unit uses pitch information. The speech processing apparatus of claim 1, wherein the speech signal converted into the frequency domain is represented by a harmonic part and a noise part as shown in Equation 1 below. [Equation 1]

 The speech processing apparatus of claim 1, wherein the noise processed signal is represented by a harmonic part and a suppressed noise part as shown in Equation 2 below. [Equation 2]

 2. The method of claim 1, wherein the two laws (constraint): 1. The signal before and after the noise processing are equal in energy, and 2. The signal after the noise processing is close to the signal before the noise processing. Characterized in that the voice processing device.  6. The speech processing apparatus of claim 5, wherein the two laws are applied to a speech signal in a vector form as in Equation 3, arranged as in Equation 4, and a noise suppression index is obtained as in Equation 5. [Equation 3]

,

[Equation 4]

[Equation 5]

In the voice processing method, Converting the input voice signal into the frequency domain when the voice signal is input, Separating a harmonic part and a noise part from the speech signal converted into the frequency domain; Determining an optimum noise suppression index k according to the situation and system, Suppressing the separated noise portion according to the noise suppression index to output a noise processed signal. 8. The method of claim 7, wherein the harmonic-noise separation step uses pitch information. The speech processing method of claim 7, wherein the speech signal converted into the frequency domain is represented by a harmonic part and a noise part as shown in Equation 1 below. [Equation 1]

 8. The speech processing method of claim 7, wherein the noise processed signal is represented by a harmonic part and a suppressed noise part as shown in Equation 2 below. [Equation 2]

 8. The method according to claim 7, wherein two constraints are obtained to obtain the noise processed signal: 1. The signal before and after the noise processing has the same energy, and 2. The signal after the noise processing is close to the signal before the noise processing. A voice processing method characterized by the above-mentioned.  12. The speech processing method of claim 11, wherein the two laws are applied to a speech signal in a vector form as in Equation 3, arranged as in Equation 4, and a noise suppression index is obtained as in Equation 5. [Equation 3]

,

[Equation 4]

[Equation 5]

In the apparatus for processing a voice signal, A voice signal input unit for receiving a voice signal, From the input voice signal, the amplification of the harmonic part and the attenuation of the noise part are repeatedly performed until the energy difference between the two consecutive harmonic components becomes equal to or less than a preset threshold value, and the energy difference between the two consecutive harmonic components is previously set. A harmonic-noise separator for separating the harmonic part and the noise part when the threshold value is less than a predetermined threshold value; A noise suppression index determination unit for determining an optimal noise suppression index k according to a situation and a system; And a noise suppression unit configured to output a noise processed signal by suppressing the separated noise portion according to the noise suppression index. The method of claim 13, wherein the harmonic-noise separation unit, A harmonic section determination unit for extracting cepstrum and pitch information to determine a harmonic section; A harmonic-noise separation repeater which repeatedly performs amplification of the harmonic part and attenuation of the noise part; The amplification and attenuation of the noise part of the harmonic couple are repeated through the harmonic-noise separation repeater until the energy difference between two consecutive harmonic components in the speech signal passing through the harmonic-noise separation repeater becomes less than or equal to a preset threshold value. Harmonic-noise separation determination unit to perform, And a harmonic-noise separation unit for separating the harmonic portion and the noise portion from the speech signal passing through the harmonic-noise separation determination unit. The method of claim 14, wherein the harmonic-noise separation repeater comprises: A harmonic extrapolation unit for setting a frequency domain value of the noise section to '0' and extrapolating current harmonic samples of the harmonic sections to the noise section; A noise judging unit which subtracts harmonic samples of the noise intervals from the initial noise sample and extrapolates the remaining noise sample estimates to the harmonic intervals; A noise extrapolation unit configured to set a frequency domain value of the harmonic interval to '0' and to extrapolate the current noise samples of the noise interval to the harmonic interval; And a harmonic determination unit which subtracts noise samples of the harmonic intervals from the initial harmonic sample and extrapolates the remaining harmonic sample estimates to the noise interval. In the voice processing method, Repeating the amplification of the harmonic part and the attenuation of the noise part until the energy difference between two consecutive harmonic components is less than a preset threshold value from the input voice signal; Separating the harmonic part and the noise part when the energy difference between two consecutive harmonic components after the amplification of the harmonic part and the attenuation of the noise part is less than or equal to a preset threshold value; Determining an optimal noise suppression index k according to the situation and the system; And suppressing the separated noise part according to the noise suppression index to output a noise processed signal. The method of claim 16, wherein the separating the harmonic portion and the noise portion comprises: A process of judging a harmonic section that determines a harmonic section by extracting cepstrum and pitch information; Performing amplification of the harmonic part and attenuation of the noise part, After amplifying the harmonic part and attenuating the noise part, checking whether an energy difference between two consecutive harmonic components in a speech signal falls below a preset threshold value; And separating the harmonic part and the noise part of the speech signal when the energy difference between two consecutive harmonic components is less than or equal to a predetermined threshold value after performing the checking process. 18. The method of claim 17, further comprising: continuing to amplify the harmonic portion and attenuate the noise portion if the energy difference between two consecutive harmonic components does not fall below a predetermined threshold value after performing the inspection process. Voice processing method. The method of claim 17, wherein the amplification of the harmonic portion and the attenuation of the noise portion are performed. Setting the frequency domain value of the noise section to '0' and performing extrapolation of current harmonic samples of the harmonic sections to the noise section, Subtracting the harmonic samples of the noise intervals from the initial noise sample, and performing extrapolation of the remaining noise sample estimates into the harmonic intervals; Setting the frequency domain value of the harmonic interval to '0' and performing extrapolation of current noise samples of the noise interval into the harmonic interval; And subtracting noise samples of the harmonic intervals from the initial harmonic sample and performing extrapolation of the remaining harmonic sample estimates into the noise interval.

Images