KR102156102B1 - Apparatus and method for noise reduction of bone conduction speech signal - Google Patents

Abstract

Disclosed are an apparatus, a method and a recording medium for removing noise of a bone conduction speech signal capable of effectively removing noise from a bone conduction speech signal by means of an entropy gate and a correlation gate. A method for removing noise from a bone conduction voice signal according to an embodiment of the present invention includes: collecting a bone conduction voice signal by a bone conduction vibration sensor; Dividing the bone conduction speech signal into a plurality of first bands by wavelet packet decomposition; Calculating energy for each of the first bands and calculating an energy threshold based on the energy calculated for each of the first bands; Determining second bands having an audio section among the plurality of first bands based on the energy threshold; Calculating entropy for each of the second bands, and generating an entropy gate based on the entropies of the second bands; Calculating a correlation of the second bands and generating a correlation gate based on the correlation of the second bands; Removing noise from the second bands by the entropy gate and the correlation gate; And generating a noise-removed bone conduction speech signal by synthesizing the noise-removed second bands.

Description

Apparatus and method for noise reduction of bone conduction speech signal

The present invention relates to a bone conduction speech signal noise removal apparatus and method, and more particularly, to a bone conduction speech signal noise removal apparatus and method capable of effectively removing noise from a bone conduction speech signal by an entropy gate and a correlation gate. About.

Recently, in various voice signal processing applications, the phenomenon of deterioration in voice recognition due to external environmental noise is recognized as an important problem to be solved, and a high-performance voice enhancement technology is required. Voice enhancement is a technique that improves the voice signal by removing noise from the voice signal and reinforcing the voice when the voice signal is input because it is contaminated by ambient noise. Communication quality of voice communication devices used during extreme work environments or military operations. It can improve speech recognition or speaker recognition performance during human-device interaction in various smart devices or medical devices such as implantable hearing aids. It can also be used in sound equipment such as headsets and digital hearing aids to suppress background noise and improve sound quality.

In order to reduce the effect of noise, various noise reduction techniques and speech enhancement techniques have been studied and are used in various speech signal processing fields. For example, a wavelet transform-based algorithm is being studied for noise removal and speech enhancement. However, conventional noise reduction algorithms have been mainly focused on removing noise from an air conduction type speech signal, and such a conventional noise reduction method may not effectively remove noise from a bone conduction type speech signal.

The bone conduction audio signal is transmitted through an auditory path different from the air conduction audio signal, and has different noise characteristics from the air conduction audio signal due to a difference in transmission medium. Bone conduction voice signals reach the cochlea through the skin and skull through vibration in specific areas such as mastoid processes and temples. Due to the difference in transmission medium, bone conduction hearing aids are more efficient in transmitting voice signals even in noisy environments than air conduction hearing aids in the same noisy environment, but in extreme environments, sound vibrations of air flow into the hearing aids through the skin. Problems arise. For this reason, when noise is removed from the bone conduction speech signal using the noise removal technology based on the air conduction speech signal, the noise is not properly removed, the loss of bone conduction speech information increases, and it is difficult to ensure speech continuity. Can cause problems.

An object of the present invention is to provide a bone conduction speech signal noise removal apparatus, method, and recording medium capable of effectively removing noise from a bone conduction speech signal through an entropy gate and a correlation gate.

The problem to be solved by the present invention is not limited to the problems mentioned above. Other technical problems that are not mentioned will be clearly understood by those of ordinary skill in the art from the following description.

A method for removing noise from a bone conduction speech signal according to an aspect of the present invention provides a method for removing noise from a bone conduction speech signal, comprising: collecting a bone conduction speech signal by a bone conduction vibration sensor; Dividing the bone conduction speech signal into a plurality of first bands by wavelet packet decomposition; Calculating energy for each of the first bands and calculating an energy threshold based on the energy calculated for each of the first bands; Determining second bands having an audio section among the plurality of first bands based on the energy threshold; Calculating entropy for each of the second bands, and generating an entropy gate based on the entropies of the second bands; Calculating a correlation of the second bands and generating a correlation gate based on the correlation of the second bands; Removing noise from the second bands by the entropy gate and the correlation gate; And generating a noise-removed bone conduction speech signal by synthesizing the noise-removed second bands.

The generating of the entropy gate may include calculating an average value of entropy of the second bands and a log value of entropy of each of the second bands; And generating the entropy gate based on the average value and the log value.

In the generating of the entropy gate, the entropy gate may be generated according to Equation 1 below.

[Equation 1]

는 상기 제2 밴드들 각각의 엔트로피, k 및 N은 상기 제2 밴드들의 개수이다.In Equation 1, EG(t) is the entropy gate,

Is the entropy of each of the second bands, and k and N is the number of the second bands.

The removing of the noise may include first removing the noise of the second bands by the entropy gate; And secondary removing noise of the second bands passing through the entropy gate by the correlation gate.

In the generating of the correlation gate, the correlation gate may be generated based on a correlation of the second bands and a second band passing through the entropy gate.

In the step of generating the correlation gate, the correlation gate may be generated according to Equation 2 below.

[Equation 2]

In Equation 2, CG(t) is the correlation gate, Ψ _ck is the k-th second band of the second bands, m is the average value of the second bands, n is the number of the second bands, Ψ _EG is a second band that has passed through the entropy gate.

In the generating of the correlation gate, the correlation gate may be generated based on a correlation of the second bands and a feature band having the highest similarity to a bone conduction voice characteristic.

In another embodiment of the present invention, the removing of the noise may include first removing noise of the second bands by the correlation gate; And secondarily removing noise of the second bands passing through the correlation gate by the entropy gate.

The dividing into the plurality of first bands may include dividing the bone conduction audio signal into a low frequency band and a high frequency band; It may include dividing the low frequency band and the high frequency band, respectively, so that the bone conduction voice signal is divided into the plurality of first bands. The feature band may include the low frequency band.

According to another aspect of the present invention, a computer-readable recording medium in which a program for executing the method for removing noise from a bone conduction voice signal is recorded is provided.

According to another aspect of the present invention, there is provided an apparatus for removing noise from a bone conduction audio signal for removing noise from a bone conduction audio signal, comprising: a bone conduction audio signal collection unit for collecting a bone conduction audio signal by a bone conduction vibration sensor; A wavelet packet decomposition unit for dividing the bone conduction speech signal into a plurality of first bands by wavelet packet decomposition; An energy threshold calculating unit that calculates energy for each of the first bands and calculates an energy threshold based on the energy calculated for each of the first bands; A band determination unit configured to determine second bands having an audio section among the plurality of first bands based on the energy threshold; An entropy gate generator for calculating entropy for each of the second bands and generating an entropy gate based on entropy of the second bands; A correlation gate generator for calculating a correlation of the second bands and generating a correlation gate based on the correlation of the second bands; A noise removing unit removing noise from the second bands by the entropy gate and the correlation gate; And a synthesizer for synthesizing the noise-removed second bands to generate a noise-removed bone conduction speech signal.

The entropy gate generator calculates an average value of entropy of the second bands and a log value of entropy of each of the second bands; In addition, the entropy gate may be generated based on the average value and the log value.

The noise removal unit first removes noise of the second bands by the entropy gate; In addition, noise of the second bands passing through the entropy gate may be secondarily removed by the correlation gate.

The correlation gate generator may generate the correlation gate based on a correlation of the second bands and a second band that has passed through the entropy gate.

The correlation gate generator may generate the correlation gate based on the correlation of the second bands and a feature band having the highest similarity to the bone conduction voice characteristic.

In another embodiment of the present invention, the noise removal unit is configured to first remove noise of the second bands by the correlation gate; In addition, noise of the second bands passing through the correlation gate may be secondarily removed by the entropy gate.

The wavelet packet decomposition unit divides the bone conduction audio signal into a low frequency band and a high frequency band; In addition, the low frequency band and the high frequency band may be respectively divided so that the bone conduction voice signal is divided into the plurality of first bands. The feature band may include the low frequency band.

The present invention provides a bone conduction speech signal noise removal apparatus, method, and recording medium capable of effectively removing noise from a bone conduction speech signal through an entropy gate and a correlation gate.

The effect of the present invention is not limited to the above-described effects. Effects not mentioned will be clearly understood by those of ordinary skill in the art from the present specification and the accompanying drawings.

1 is a conceptual diagram of a method for removing noise from a bone conduction speech signal according to an embodiment of the present invention.
2 is a flowchart of a method for removing noise from a bone conduction speech signal according to an embodiment of the present invention.
3 is a block diagram of an apparatus for removing noise from a bone conduction speech signal according to an embodiment of the present invention.
4 is a conceptual diagram illustrating a wavelet packet decomposition of a bone conduction voice signal according to an embodiment of the present invention.
5 is a block diagram of a voice enhancement unit constituting an apparatus for removing noise from a bone conduction voice signal according to an embodiment of the present invention.
6 is a schematic diagram of an experimental model for evaluating the performance of an apparatus for removing noise from a bone conduction speech signal according to an embodiment of the present invention.
7 is a view showing a noise removal result of the bone conduction speech signal noise removal apparatus according to an embodiment of the present invention.
8 is a spectral result of a bone conduction speech signal from which noise is removed according to an embodiment of the present invention.
9 is a result of comparing the noise removal performance of the bone conduction speech signal noise removal method according to an embodiment of the present invention with a conventional noise removal method.

Other advantages and features of the present invention, and a method of achieving them will become apparent with reference to embodiments to be described later in detail together with 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. Even if not defined, all terms (including technical or scientific terms) used herein have the same meaning as commonly accepted by universal technology in the prior art to which this invention belongs. General descriptions of known configurations may be omitted so as not to obscure the subject matter of the present invention. In the drawings of the present invention, the same reference numerals are used as much as possible for the same or corresponding configurations. In order to help the understanding of the present invention, some configurations in the drawings may be somewhat exaggerated or reduced.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present application, terms such as "comprise", "have" or "have" are intended to designate the presence of features, numbers, steps, actions, components, parts, or a combination thereof described in the specification. It is to be understood that the possibility of the presence or addition of other features, numbers, steps, actions, components, parts, or combinations thereof, or any further features, is not excluded in advance.

The'~ unit' used throughout this specification is a unit that processes at least one function or operation, and may mean, for example, a hardware component such as software, FPGA, or ASIC. However,'~ part' is not limited to software or hardware. The'~ unit' may be configured to be in an addressable storage medium, or may be configured to reproduce one or more processors.

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

1 is a conceptual diagram of a method for removing noise from a bone conduction speech signal according to an embodiment of the present invention. Referring to FIG. 1, in the method for removing noise from a bone conduction speech signal according to an embodiment of the present invention, after filtering a bone conduction speech signal mixed with noise, it is divided into first bands by wavelet packet decomposition. In steps S1 to S3, energy is calculated for each first band to calculate an energy threshold, and based on the energy threshold, second bands having a negative section among the first bands are determined, and then entropy for each of the second bands. The step of generating an entropy gate (entropy window) by calculating (S4), the step of generating a correlation gate (correlation window) based on the correlation of the second bands (S5), and the entropy gate and the correlation gate And removing noise from the second bands (S6), and generating a noise-removed bone conduction speech signal by synthesizing (recombining) the second bands from which the noise has been removed (S7 to S8).

2 is a flowchart of a method for removing noise from a bone conduction speech signal according to an embodiment of the present invention. 3 is a block diagram of an apparatus for removing noise from a bone conduction speech signal according to an embodiment of the present invention. Referring to FIGS. 2 and 3, the apparatus 100 for removing noise from a bone conduction speech signal includes a bone conduction speech signal collection unit 120, a wavelet packet decomposition unit 140, and a speech enhancement unit 160.

The bone conduction audio signal collection unit 120 collects a bone conduction audio signal including noise (S10). In one embodiment, the bone conduction voice signal collection unit 120 may include a bone conduction vibration sensor for collecting a bone conduction voice signal and/or a device for receiving a bone conduction voice signal collected by the bone conduction vibration sensor. have. The bone conduction audio signal y(n) mixed with noise can be expressed as Equation (1) below.

[Equation (1)]

In Equation (1), s ( n ) is a clean bone conduction speech signal without noise of the nth frame, and v ( n ) is the background noise of the nth frame. When a bone conduction audio signal mixed with noise is input from the bone conduction audio signal collection unit 120, the bone conduction audio signal is filtered by, for example, a 100 Hz high pass filter, and then the signal section and the noise section are extracted (distinguished). To do this, wavelet packet decomposition is performed by the wavelet packet decomposition unit 140 (S40). Accordingly, the bone conduction audio signal is divided into first bands in a time-frequency two-dimensional region.

In one embodiment, the wavelet packet decomposition unit 140 is not simply based on an arithmetic band-specific energy, but based on a human acoustic and auditory model, according to the amount of energy stimulated by the human auditory nerve, and converts the speech signal into FIG. It can be decomposed into 20 first bands as shown in FIG. The first bands decomposed from the bone conduction speech signal have a form of a time domain signal having 20 frequency information, and can represent both time and frequency information in a two-dimensional matrix.

In one embodiment, the speech signal may be decomposed into wavelet coefficients ( w _j,m ( k )) having 20 subbands. Since the bone conduction audio signal is mainly concentrated in the low frequency band, the number of subbands in the low frequency band is larger than the number of subbands in the high frequency band. The wavelet coefficient w _j,m (k) represents the k-th wavelet coefficient of the j- th level and m- th wavelet band ( j =2,3,4, m =1,...,20). The bone conduction audio signal can be decomposed into 20 bands at the fourth level. In order to process the wavelet coefficients w _j,m (k) simultaneously with information in the time and frequency domains, the wavelet coefficients can be expressed as a two-dimensional matrix as shown in Equation (2) below.

[Equation (2)]

은 특정시간 t에서의 m번째 제1 밴드(서브밴드)의 웨이블릿 계수의 행렬로 이루어진다. 다시 도 2 및 도 3을 참조하면, 웨이블릿 패킷 분해부(140)에 의해 골전도 음성신호가 다수의 제1 밴드들(서브밴드들)로 분해되면, 음성 향상부(160)는 제1 밴드들 별로 잡음을 제거한 후, 잡음 제거된 밴드들을 합성하여 음성 향상된 골전도 음성신호를 생성한다.In equation (2),

Is composed of a matrix of wavelet coefficients of the m-th first band (subband) at a specific time t. Referring back to FIGS. 2 and 3, when the bone conduction voice signal is decomposed into a plurality of first bands (subbands) by the wavelet packet decomposing unit 140, the voice enhancing unit 160 includes the first bands. After each noise is removed, the noise-removed bands are synthesized to generate a voice-enhanced bone conduction voice signal.

5 is a block diagram of a voice enhancement unit constituting an apparatus for removing noise from a bone conduction voice signal according to an embodiment of the present invention. Referring to FIG. 5, the voice enhancement unit 160 includes an energy threshold calculation unit 161, a band determination unit 162, an entropy gate generation unit 163, a correlation gate generation unit 164, and a noise removal unit 165. ) And a synthesis unit 166 may be included.

2 and 5, the energy threshold calculation unit 161 calculates energy for each of the first bands (eg, 20 subbands) decomposed by the wavelet packet decomposition unit 140, and the first An energy threshold is calculated based on the energy calculated for each band (S30). The energy threshold calculation unit 161 may calculate an energy threshold ψ _th (t) for distinguishing a voice signal and a noise signal according to Equation (3) below.

[Equation (3)]

는 음성신호,

는 잡음신호, m은 제1 밴드의 순번, ψ_m(t)는 m번째 제1 밴드이다. 골전도 음성신호는 수식 (3)에 의해 완전히 잡음이 제거되지 않는다. 잡음의 효과적인 제거를 위하여, 먼저 밴드 결정부(162)는 에너지 문턱치 ψ_th(t)를 기반으로 다수의 제1 밴드들 중 음성 구간을 가지는 제2 밴드들을 결정할 수 있다(S40). 실시예에서, 20개의 제1 밴드들 중에서 음성신호를 포함하고 있는 제1 밴드들이 제2 밴드들로 결정될 수 있다. 다른 실시예에서, 음성신호를 가지고 있는 제1 밴드들 중에서 음성신호의 정보량이 설정치 이상 포함된 제1 밴드들 만을 제2 밴드들로 결정하는 것도 가능하다.In formula (3),

Is the voice signal,

Is the noise signal, m is the sequence number of the first band, and ψ _m (t) is the m-th first band. The bone conduction audio signal is not completely removed by Equation (3). In order to effectively remove noise, the band determiner 162 may first determine second bands having a speech section among a plurality of first bands based on the energy threshold ψ _th (t) (S40). In an embodiment, among the 20 first bands, first bands including an audio signal may be determined as second bands. In another embodiment, it is possible to determine only the first bands in which the information amount of the voice signal is greater than or equal to the set value among the first bands having the voice signal as the second bands.

When the second bands are determined by the band determiner 162, the entropy gate generator 163 may calculate entropy for each of the second bands and generate an entropy gate based on the entropy of the second bands. (S50). In an embodiment, the entropy gate generator 163 calculates an average value of entropy of the second bands and a log value of entropy of each of the second bands, and the average value of the calculated entropy of the second bands and each of the second bands An entropy gate can be created based on the logarithm of entropy. The entropy gate generator 163 may generate an entropy gate according to Equation (4) below.

[Equation (4)]

는 제2 밴드들 각각의 샤논 엔트로피(Shannon entropy), N(=k)은 제2 밴드들의 개수(N≤20)이다. 골전도 음성정보는 잡음보다 높은 엔트로피 값을 가지며, 엔트로피 게이트 EG(t)에 의해 1차로 잡음이 제거될 수 있다. 골전도 음성신호의 음성 향상은 엔트로피 게이트 만으로는 목표로 하는 수준에 도달하기 어렵다.In Equation (4), EG(t) is an entropy gate based on an entropy band threshold,

Is Shannon entropy of each of the second bands, and N(=k) is the number of second bands (N≦20). The bone conduction speech information has an entropy value higher than that of noise, and noise may be firstly removed by the entropy gate EG(t). It is difficult to achieve the target level of the bone conduction voice signal with only the entropy gate.

In order to maximize the speech enhancement effect of the bone conduction speech signal, the correlation gate generator 164 calculates cross-correlation of the second bands, and generates a correlation gate based on the correlation of the second bands. Generate (S60). In an embodiment, the correlation gate generator 164 may generate a correlation gate based on the correlation of the second bands and the second band that has passed through the entropy gate. The correlation gate generator 164 may generate the correlation gate CG(t) according to Equation (5) below, for example.

[Equation (5)]

In Equation (5), Ψ _ck is the k-th second band among the second bands, m is the average value of the second bands, n is the number of second bands, and Ψ _EG is the second band passing through the entropy gate.

The noise removing unit 165 may remove noise from the second bands by the entropy gate and the correlation gate (S70). In an embodiment, the noise removal unit 165 may first remove the noise of the second bands by the entropy gate first, and then remove the noise of the second bands passing through the entropy gate by the correlation gate. .

으로 치환될 수 있으며, 수식 (4)에서

는 상관관계 게이트를 통과한 제2 밴드들 각각의 샤논 엔트로피(Shannon entropy)일 수 있다.

(도 4 참조)은 골전도 음성신호로부터 첫번째 레벨에서 분할된 저주파수 대역 및 고주파수 대역 중 저주파수 대역을 포함할 수 있다.Alternatively, the noise removal unit 165 may first remove the noise of the second bands by the correlation gate and then remove the noise of the second bands passing through the correlation gate by the entropy gate. The correlation gate generator 164 may generate a correlation gate based on a correlation between the second bands and a feature band having the highest similarity to the bone conduction voice characteristic. In this case, in Equation (5), Ψ _EG has characteristics very similar to the bone conduction audio signal.

Can be substituted with, in Equation (4)

May be Shannon entropy of each of the second bands passing through the correlation gate.

(See FIG. 4) may include a low frequency band of a low frequency band and a high frequency band divided at the first level from the bone conduction voice signal.

When the second bands of the bone conduction speech signal are removed from the first and second noises based on the entropy gate and the correlation gate by the noise removal unit 165, the synthesis unit 166 reverses the noise removed second bands. A bone conduction speech signal from which noise is removed may be generated by synthesizing by inverse wavelet packet decomposition.

According to an apparatus and method for removing noise from a bone conduction voice signal according to an embodiment of the present invention, external environmental noise that contaminates the voice signal in the process of transmitting the vibration of sound through the air from the bone conduction hearing aid to the bone conduction hearing aid through the skin. It can be effectively removed, and efficient voice signal transmission is possible. An experiment was conducted to evaluate the performance of the bone conduction speech signal noise removal method according to an embodiment of the present invention. 6 is a schematic diagram of an experimental model for evaluating the performance of an apparatus for removing noise from a bone conduction speech signal according to an embodiment of the present invention. An electric condenser microphone (ECM) and a piezo-electric sensor were used in the experimental model.

The electric condenser microphone was placed close to the experimenter's mouth to collect the air conduction audio signal, and the piezo sensor was fixed to the experimenter's head to be suitable for collecting the bone conduction audio signal. A 100dB SPL speaker was used to simulate a noisy environment. The vowel'a' of 70-75dB SPL was pronounced repeatedly several times. The experiment was conducted in an environment of about 30-35 dB SPL background noise.

7 is a view showing a noise removal result of the bone conduction speech signal noise removal apparatus according to an embodiment of the present invention. 7A is a signal before noise removal, the upper part is the air conduction voice signal before noise removal collected by the microphone, and the lower part is the bone conduction voice signal before noise removal collected by the piezo sensor. 7B is a signal after noise is removed by an entropy gate and a correlation gate. The upper part is an air conduction voice signal after noise removal, and the lower part is a bone conduction voice signal after noise removal. The vowel'a' was repeatedly pronounced, and white noise and crosstalk noise occurred from 2.5 seconds.

It can be seen that noise is efficiently removed by entropy and correlation thresholds according to an embodiment of the present invention. From the upper voice waveform of FIG. 7 (b), in the case of the air-conducting voice signal, due to the propagation characteristics different from the bone-conducting voice signal, the first and second'a' voice information was lost along with the removal of white noise, It can be seen that noise is not effectively removed from the 7th'a' voice information.

8 is a spectral result of the bone conduction audio signal from which noise is removed according to an embodiment of the present invention. FIG. 8A is a voice signal having crosstalk noise, (b) a voice signal having white noise, (c) is a voice signal from which crosstalk noise is removed according to an embodiment of the present invention, and (d) is a spectral result of a voice signal from which white noise is removed according to an embodiment of the present invention. Crosstalk noise (babble noise) has a characteristic very similar to that of a speech signal, so it is difficult to distinguish it from a speech signal by a conventional speech enhancement method, but the bone conduction speech signal noise removal method according to an embodiment of the present invention includes white noise and crosstalk. It can be seen that it exhibits excellent noise reduction performance against noise.

9 is a result of comparing the noise removal performance of the bone conduction speech signal noise removal method according to an embodiment of the present invention with a conventional noise removal method. 9A is a speech signal and spectrum noise-removed by a conventional adaptive filter, (b) a speech signal and spectrum noise-removed by a conventional spectral subtraction, (c) ) Shows a noise-removed speech signal and spectrum according to an embodiment of the present invention. From the spectra shown in FIGS. 9A and 9B, it can be seen that in the case of the conventional noise removal method, noise is not completely removed. On the other hand, as shown in (c) of FIG. 9, in the case of the bone conduction speech signal noise removal method according to an embodiment of the present invention, it can be seen that noise is cleanly removed in both the time domain and the frequency domain.

The method according to an embodiment of the present invention can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. Computer-readable recording media include volatile memories such as SRAM (Static RAM), DRAM (Dynamic RAM), SDRAM (Synchronous DRAM), Read Only Memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Nonvolatile memory such as Electrically Erasable and Programmable ROM (EEPROM), flash memory device, phase-change RAM (PRAM), magnetic RAM (MRAM), resistive RAM (RRAM), ferroelectric RAM (FRAM), floppy disk, hard disk, or The optical reading medium may be, for example, a storage medium in the form of a CD-ROM or a DVD, but is not limited thereto.

It should be understood that the above embodiments have been presented to aid the understanding of the present invention, and do not limit the scope of the present invention, and various deformable embodiments are also within the scope of the present invention. The technical protection scope of the present invention should be determined by the technical spirit of the claims, and the technical protection scope of the present invention is not limited to the literal description of the claims per se, but the invention of the scope of which the technical value is substantially equal. It should be understood that it reaches to.

100: bone conduction speech signal noise removal device
120: bone conduction voice signal collection unit
140: wavelet packet decomposition unit
160: voice enhancement unit
161: energy threshold calculation unit
162: band determination unit
163: entropy gate generation unit
164: correlation gate generation unit
165: noise canceling unit
166: synthesis unit

Claims (17)

In the bone conduction speech signal noise removal method for removing noise from the bone conduction speech signal,
Collecting a bone conduction voice signal by a bone conduction vibration sensor;
Dividing the bone conduction speech signal into a plurality of first bands by wavelet packet decomposition;
Calculating energy for each of the first bands and calculating an energy threshold based on the energy calculated for each of the first bands;
Determining second bands having an audio section among the plurality of first bands based on the energy threshold;
Calculating entropy for each of the second bands, and generating an entropy gate based on the entropies of the second bands;
Calculating a correlation of the second bands and generating a correlation gate based on the correlation of the second bands;
Removing noise from the second bands by the entropy gate and the correlation gate; And
Comprising the step of synthesizing the noise-removed second bands to generate a noise-removed bone conduction speech signal,
Generating the entropy gate,
Calculating an average value of entropy of the second bands and a log value of entropy of each of the second bands; And
And generating the entropy gate based on the average value and the log value. delete The method of claim 1,
In the step of generating the entropy gate, the entropy gate is generated according to Equation 1 below,
[Equation 1]

In Equation 1, EG(t) is the entropy gate,

Is the entropy of each of the second bands, and k and N are the number of the second bands. The method of claim 3,
The step of removing the noise,
First removing noise of the second bands by the entropy gate; And
And secondly removing noise of the second bands passing through the entropy gate by the correlation gate. The method of claim 4,
The generating of the correlation gate comprises generating the correlation gate based on a correlation of the second bands and a second band passing through the entropy gate. The method of claim 5,
In the step of generating the correlation gate, generating the correlation gate according to Equation 2 below,
[Equation 2]

In Equation 2, CG(t) is the correlation gate, Ψ _ck is the k-th second band of the second bands, m is the average value of the second bands, n is the number of the second bands, Ψ _EG is the second band passing through the entropy gate, bone conduction speech signal noise removal method. The method of claim 1,
The generating of the correlation gate comprises generating the correlation gate based on the correlation of the second bands and a feature band having the highest similarity to the bone conduction voice feature. The method of claim 7,
The step of removing the noise,
First removing noise of the second bands by the correlation gate; And
And removing, by the entropy gate, the noise of the second bands passing through the correlation gate by a second order. The method of claim 8,
The step of dividing into the plurality of first bands,
Dividing the bone conduction audio signal into a low frequency band and a high frequency band;
Dividing the low frequency band and the high frequency band, respectively, so that the bone conduction audio signal is divided into the plurality of first bands,
The characteristic band is a bone conduction speech signal noise removal method including the low frequency band. A computer-readable recording medium in which a program for executing the method for removing noise of a bone conduction speech signal according to any one of claims 1, 3 to 9 is recorded. In the bone conduction speech signal noise removal apparatus for removing noise from the bone conduction speech signal,
Bone conduction voice signal collection unit for collecting the bone conduction voice signal by the bone conduction vibration sensor;
A wavelet packet decomposition unit for dividing the bone conduction speech signal into a plurality of first bands by wavelet packet decomposition;
An energy threshold calculating unit that calculates energy for each of the first bands and calculates an energy threshold based on the energy calculated for each of the first bands;
A band determination unit configured to determine second bands having an audio section among the plurality of first bands based on the energy threshold;
An entropy gate generator for calculating entropy for each of the second bands and generating an entropy gate based on entropy of the second bands;
A correlation gate generator for calculating a correlation of the second bands and generating a correlation gate based on the correlation of the second bands;
A noise removing unit removing noise from the second bands by the entropy gate and the correlation gate; And
A synthesizer for synthesizing the noise-removed second bands to generate a noise-removed bone conduction speech signal,
The entropy gate generation unit,
Calculating an average value of entropy of the second bands and a log value of entropy of each of the second bands; And
Bone conduction speech signal noise removal apparatus for generating the entropy gate based on the average value and the log value. delete The method of claim 11,
The noise removal unit,
First-order noise cancellation of the second bands by the entropy gate; And
Bone conduction speech signal noise removal apparatus for secondarily removing noise of second bands passing through the entropy gate by the correlation gate. The method of claim 13,
The correlation gate generator generates the correlation gate based on the correlation of the second bands and the second band passing through the entropy gate. The method of claim 11,
The correlation gate generating unit generates the correlation gate based on the correlation of the second bands and the characteristic band having the highest similarity to the bone conduction speech characteristic. The method of claim 15,
The noise removal unit,
First-order removal of noise in the second bands by the correlation gate; And
Bone conduction speech signal noise removal apparatus for secondary removal of noise of the second bands passing through the correlation gate by the entropy gate. The method of claim 16,
The wavelet packet decomposition unit,
Dividing the bone conduction audio signal into a low frequency band and a high frequency band; And
Dividing the low frequency band and the high frequency band, respectively, so that the bone conduction audio signal is divided into the plurality of first bands,
The characteristic band is a bone conduction speech signal noise removal apparatus including the low frequency band.

Images