KR101098763B1 - Method and system of suppressing noise - Google Patents

Abstract

The present invention relates to a noise removing method, comprising: determining a plurality of frames included in an input signal having a noise signal and a voice signal as a noise section or a voice section, and at least one noise frame determined as a noise section among the plurality of frames Noise is estimated by using a noise estimated from a predetermined number of noise frames for a voice transition frame transitioned from a noise period to a voice interval among a plurality of frames as an estimated noise for the voice transition frame. And remove noise by using estimated noise in at least one voice frame determined as a voice section among the plurality of frames.

Description

Method and system of suppressing noise

The present invention relates to voice / audio signal processing among multimedia signals, and more particularly, to a method and a system for removing noise included in a voice / audio signal.

In the mobile communication field, multimedia signal processing technology refers to processing various information by recognizing various expression media such as sound or video. With the advent of digital mobile communication systems and the development of wired communication systems, high-level voice / audio codec technology has been developed. Important factors in the voice / audio codec technology include transmission speed, sound quality under various environments, voice encoding delay time, and complexity, among which sound quality is particularly important in practical applications.

Recently, a technology for making a voice call using a mobile communication terminal has been generally used. However, when a user makes a call to such a mobile communication terminal in a train, street, or music venue where there is a loud noise of the surrounding environment, the call quality or playback sound quality may be degraded due to the excessive noise. have. In this case, a technique of removing background noise from a signal input to the terminal is required. When the energy level of the voice signal is low, the voice signal may be buried in the background noise due to the background noise.

The problem to be solved by the present invention is a noise reduction method and system that can improve the quality of the speech signal output from the input signal by effectively estimating the noise in the noise section of the input signal having a noise signal and the voice signal, and the noise A computer readable recording medium having recorded thereon a program for executing the removal method is provided.

According to an aspect of the present invention, there is provided a noise removing method, comprising: determining a plurality of frames included in an input signal having a noise signal and a voice signal as a noise section or a voice section; Estimating noise in at least one noise frame determined as the noise section among the plurality of frames; Noise in the speech transition frame is estimated using noise estimated from a noise frame before the predetermined number of speech transition frames transitioned from the noise section to the speech section among the plurality of frames as the estimated noise for the speech transition frame. Removing; And removing noise from the at least one speech frame determined as the speech section among the plurality of frames by using the estimated noise.

In addition, the task is to determine a plurality of frames included in the input signal having a noise signal and a voice signal as a noise interval or a voice interval; Estimating noise in at least one noise frame determined as the noise section among the plurality of frames; Noise in the speech transition frame is estimated using noise estimated from a noise frame before the predetermined number of speech transition frames transitioned from the noise section to the speech section among the plurality of frames as the estimated noise for the speech transition frame. Removing; And removing noise from the at least one voice frame determined as the voice interval among the plurality of frames using the estimated noise. 2. A computer-readable recording medium having recorded thereon a program for executing the noise removing method. Is achieved by.

In addition, a noise canceling system according to the present invention for solving the above problems is a voice / noise section determination unit for discriminating a plurality of frames included in the input signal having a noise signal and a voice signal as a noise section or a speech section; A transition determining unit determining whether a transition has occurred in at least one noise frame determined as the noise section or at least one voice frame determined as the speech section among the plurality of frames; Estimating noise in the at least one noise frame, and updating noise estimated in a predetermined number of noise frames with respect to the speech transition frame to the speech transition frame in which the transition from the noise section to the speech section occurs. A noise estimation and updating unit; And a noise removing unit for removing noise from the speech transition frame or the speech frame using the estimated noise.

According to the present invention, by removing the noise from the speech transition frame using the noise estimated from the noise frame before the predetermined number for the speech transition frame transitioned from the noise interval to the speech interval as the estimated noise for the speech transition frame, Although the speech signal is included in the noise frame determined as, the noise estimated in the noise frames is not used in a subsequent noise removal operation. As a result, when the noise removing operation is performed, the noise component removed from the input signal includes almost no voice signal component, thereby preventing distortion of the output voice signal.

In addition, since the noise estimation operation is not performed on a noise transition frame after a predetermined number of noise transition frames transitioned from the speech section to the noise section, even if the noise signal determined as the noise section includes the voice signal, the corresponding noise frames are the subsequent noise. Does not affect the removal operation.

1 is a block diagram schematically illustrating a voice / audio codec according to an embodiment of the present invention.
2 is a block diagram illustrating a noise cancellation system according to an exemplary embodiment of the present invention.
3 is a graph illustrating a noise section and a voice section of an input signal input to the noise removing system of FIG. 2.
4 is a graph illustrating an example of energy levels of a noise signal, a voice signal, and an input signal input to the noise removing system of FIG. 2.
FIG. 5 illustrates an example in which frames included in an input signal transition from a noise section to a voice section as an output of the voice / noise section discriminating unit of FIG. 2.
FIG. 6 illustrates an example in which frames included in an input signal are transitioned from a speech section to a noise section as an output of the speech / noise section discriminator of FIG. 2.
7A and 7B schematically illustrate the configuration of the estimated noise storage unit of FIG. 2 when a is 4.
8 is a flowchart illustrating a noise removing method according to an embodiment of the present invention.

With respect to the embodiments of the present invention disclosed in the text, specific structural to functional descriptions are merely illustrated for the purpose of describing embodiments of the present invention, embodiments of the present invention may be implemented in various forms and It should not be construed as limited to the embodiments described in.

Hereinafter, with reference to the accompanying drawings, it will be described in detail a preferred embodiment of the present invention. The same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

1 is a block diagram schematically illustrating a voice / audio codec according to an embodiment of the present invention.

Referring to FIG. 1, the voice / audio codec 10 includes an input signal processor 11, a voice / audio processor 12, and an output signal processor 13. Here, the voice / audio codec 10 may be an encoder or a decoder.

The input signal processor 11 may include an analog to digital converter, and may convert an analog input signal into digital. The audio / audio processor 12 may encode / decode the digital signal output from the input signal processor 11 in units of frames. The output signal processor 13 may include a digital to analog converter, and may convert a digital signal output from the voice / audio processor 12 into an analog signal.

In the present embodiment, the input signal processor 11 or the output signal processor 13 may further include a noise cancellation system. In detail, when the voice / audio codec 10 is an encoder, the input signal processor 11 may include a noise removing system. When the voice / audio codec 10 is a decoder, the output signal processor 13 may remove noise. A system may be included, but the exemplary embodiment is not limited thereto, and both the input signal processor 11 and the output signal processor 13 may include a noise canceling system.

The voice / audio codec 10 may be a computer terminal, a public switched telephone network (PSTN) terminal, VoIP, a session initiation protocol (SIP), a megaco, a personal digital assistant (PDA), a cellular phone, a personal communication service (PCS) phone. , Hand-Held PC, CDMA-2000 (1X, 3X) Phone, Wideband CDMA Phone, Dual Band / Dual Mode Phone, Global Standard for Mobile (GSM) Phone It may be included in a mobile broadband system (MBS) phone, or a satellite / terrestrial digital multimedia broadcasting (DMB) phone, a recorder, or a black box.

2 is a block diagram illustrating a noise cancellation system according to an exemplary embodiment of the present invention.

Referring to FIG. 2, the noise removing system 20 includes a speech / noise section discriminating unit 21, a transition determining unit 22, a noise estimating and updating unit 23, an estimated noise storing unit 24, and a converting unit ( 25) and the noise canceller 26 may be included.

The voice / noise section determination unit 21 receives an input signal IN having a noise signal and a voice signal, and converts a plurality of frames included in the received input signal IN into a noise section or a voice section. Can be determined. In the present embodiment, the voice / noise section determination unit 21 may be implemented with Voice Activity Detection (VAD), and whether the packet is generated by detecting whether the amplitude value of the voice signal reaches a predetermined threshold. By determining, the plurality of frames may be determined as a noise section or a voice section.

In general, since a certain amount of background noise, i.e., a noise signal, exists in a call environment or the like, when a voice signal is generated, an input signal including the noise signal and the voice signal is input to the voice / noise section discriminator 21. Here, the noise section is a section including only the noise signal, not the voice signal, the speech section is a section including the noise signal and the voice signal.

3 is a graph illustrating a noise section and a voice section of an input signal input to the noise removing system of FIG. 2.

Referring to FIG. 3, the X axis represents time and the Y axis represents energy level. The voice / noise section discriminating unit 21 analyzes the energy level of the input signal IN and discriminates the noise section when the amount of change in the energy level is within a certain range, and the voice when the amount of change in the energy level is above a certain range. It can be determined as a section.

When the input signal IN transitions from the noise section to the voice section, the voice / noise section discriminating unit 21 may determine the corresponding frames as the noise section despite the presence of the voice signal in a few previous frames. . In addition, when the input signal IN transitions from the speech section to the noise section, the speech / noise section discriminating unit 21 may determine the frames as the noise section despite the presence of the speech signal in several frames thereafter. Can be.

4 is a graph illustrating an example of energy levels of a noise signal, a voice signal, and an input signal input to the noise removing system of FIG. 2.

Referring to FIG. 4, the X axis represents time and the Y axis represents energy level. While the noise signal has a constant energy level in which the amount of change in energy level is not large regardless of time, the voice signal has a large amount of change in energy level over time, and thus an input signal corresponding to the sum of the noise signal and the voice signal. The energy level varies greatly with time.

Since the energy level of the voice signal is smaller than the energy level of the noise signal in the first section 41, the energy level of the input signal is greater than the energy level of the noise signal. 41 may be determined as a voice interval. On the other hand, since the energy level of the voice signal is much smaller than the energy level of the noise signal in the second to fourth sections 42, 43, and 44, the energy level of the input signal is almost similar to that of the noise signal. Accordingly, the voice / noise section discriminating unit 21 detects the second to fourth sections 42, 43, and 44 in the noise section despite the presence of the voice signal in the second to fourth sections 42, 43, and 44. Can be determined by

In particular, when the input signal IN transitions from the noise section to the voice section or when the voice section transitions to the noise section, the energy level of the voice signal may be smaller than the energy level of the input signal. In this case, the voice / noise section determination unit 21 may determine the noise section even though a voice signal exists in the input signal. Accordingly, since the noise estimating and updating unit 23 estimates the noise in the noise frame in which the speech signal exists, the speech signal is included in the estimated noise. Subsequently, the noise removing unit 26 outputs the speech signal by removing the estimated noise from the input signal. At this time, since a component similar to the speech signal may be removed from the input signal, the speech signal is damaged and the degraded speech signal is output. Can be.

Referring back to FIG. 2, the transition determiner 22 transitions from at least one noise frame determined as a noise section or a speech frame determined as a noise section among a plurality of frames included in the input signal IN. Judge whether or not has occurred. Specifically, the transition determining unit 22 determines whether the previous frame is a noise frame or a voice frame when the current frame is a noise frame, and determines that no transition has occurred in the current frame when the previous frame is a noise frame. If the frame is an audio frame, it is determined that a transition has occurred in the current frame. Meanwhile, the transition determining unit 22 determines whether the previous frame is a voice frame or a noise frame when the current frame is a voice frame, and determines that no transition has occurred in the current frame when the previous frame is a voice frame. In the case of this noise frame, it is determined that a transition has occurred in the current frame.

The transition determining unit 22 provides an output of the transition determining unit 22 to the noise estimation and updating unit 23 when the current frame is a noise frame or an audio transition frame in which a transition is made from a noise frame to a speech frame. In addition, when the current frame is an audio frame, the transition determiner 22 provides an output of the transition determiner 22 to the noise canceller 26.

Meanwhile, when the current frame is a noise transition frame in which a transition from a voice frame to a noise frame occurs, the transition determination unit 22 does not provide the noise transition frame to the noise estimation and update unit 20, and transitions in the next frame. Determine if it has occurred. In this case, the transition determining unit 22 does not provide an output to the noise estimating and updating unit 23 so that the noise estimation is not performed for several frames after the noise transition frame, or the noise estimating and updating unit 23 does not provide an output. The operation can be controlled so that it is not activated. As such, by not performing noise estimation on the next few frames of the noise transition frame, even if the voice signal is included in the noise frame, the noise included in the noise frame can be excluded from the subsequent noise removal operation.

FIG. 5 illustrates an example in which frames included in an input signal transition from a noise section to a voice section as an output of the voice / noise section discriminating unit of FIG. 2.

Referring to FIG. 5, the voice / noise section determination unit 21 analyzes a plurality of frames included in the input signal IN, and selects (n-5) th to (n-1) th frames as the noise section ( N) and the nth frame can be determined as the voice section (S). In this case, the transition determining unit 22 determines that a transition from the noise section N to the voice section S occurs in the nth frame.

FIG. 6 illustrates an example in which frames included in an input signal are transitioned from a speech section to a noise section as an output of the speech / noise section discriminator of FIG. 2.

Referring to FIG. 6, the voice / noise section determination unit 21 analyzes a plurality of frames included in the input signal IN to convert the nth frame into the voice section S and the (n + 1) th The frame to the (n + 5) th frame may be determined as the noise period N. At this time, the transition determining unit 22 determines that a transition from the voice section S to the noise section N occurs in the (n + 1) th frame.

Referring back to FIG. 2, when the noise frame is input, the noise estimating and updating unit 23 estimates the noise in the noise frame, and provides the estimated noise to the estimated noise storage unit 24. In addition, when the voice transition frame is input, the noise estimating and updating unit 23 loads the noise estimated from the noise frame before the predetermined number with respect to the voice transition frame from the estimated noise storage unit 24, and loads the loaded noise into the voice. Update to estimated noise for the transition frame. Here, the estimated noise for the speech transition frame represents a noise component used to remove noise in the speech transition frame. Specifically, when the current frame is the nth frame, the noise estimating and updating unit 23 loads the noise estimated in the (na) th frame from the estimated noise storage unit 24, and loads the loaded noise into the nth frame. Update to estimated noise for

The estimated noise storage unit 24 stores the noise estimated by the noise estimation and update unit 23. In this case, the estimated noise storage unit 24 may be implemented as a first in first out (FIFO) buffer. Specifically, the estimated noise storage 24 stores the noise components estimated in the frame (n-1) to the frame (n-a). In this case, a represents a delayed frame index, and the value of a is determined in consideration of the size of each frame and the delay period of the noise reduction system 20.

7A and 7B schematically illustrate the configuration of the estimated noise storage unit of FIG. 2 when a value is 4.

7A and 7B, since the a value is 4, the estimated noise storage 24 may store the estimated noise in four frames. When the noise estimated in the (n-2) th frame is input, the estimated noise storage unit 24 may store the noise estimated in the (n-5) th frame to the (n-2) th frame. Subsequently, when the noise estimated in the (n-1) th frame is input, the estimated noise storage unit 24 may store the noise estimated in the (n-4) th frame to the (n-1) th frame. have.

As such, the estimated noise storage unit 24 may store the estimated noise provided from the noise estimation and update unit 23 according to the FIFO method. Specifically, when the noise estimated in the (n-1) th frame is input to the estimated noise storage 24, the (n-4) th in the space where the estimated noise in the (n-5) th frame was stored. The noise estimated in the frame is stored, the noise estimated in the (n-3) th frame is stored in the space where the noise estimated in the (n-4) th frame was stored, and estimated in the (n-3) th frame. The noise estimated in frame (n-2) is stored in the space where the noise was stored, and the noise estimated in frame (n-1) is stored in the space in which the noise estimated in the (n-2) frame was stored. do.

Referring back to FIG. 2, if the nth frame is a voice transition frame and a value is 4, the noise estimating and updating unit 23 estimates the noise estimated at the (n-4) th frame stored in the estimated noise storage unit 24. And load the estimated noise in the loaded (n-4) th frame to the estimated noise required for the noise canceling operation for the nth frame. Thus, the noise estimated in the (n-1) th frame, the (n-2) th frame, and the (n-3) th frame, that is, the previous frame of the voice transition frame in which the transition from the noise interval to the speech interval occurs, is performed by the subsequent noise. Excluded from the remove operation. Therefore, even when the transition from the noise section to the speech section occurs, even if it is determined that the noise frame is included in the previous few frames, the noise estimated in the noise frame is not used in the subsequent noise canceling operation. It is possible to prevent distortion of the voice signal output to the.

The converter 25 converts the input signal IN from the time domain to the frequency domain. In this case, the converter 25 may be implemented to include a plurality of filterbanks. In detail, the converter 25 applies a window to the input signal IN, performs a modified discrete cosine transform (MDCT) on the signal to which the window is applied, and converts the input signal IN of the time domain into the frequency domain. To spectral data.

The noise canceller 26 outputs a speech signal by removing estimated noise from the speech transition frame or the converted input signal IN corresponding to the speech frame. In detail, the noise removing unit 26 may output a voice signal by subtracting a spectrum corresponding to the estimated noise from the input signal IN converted into the frequency domain.

8 is a flowchart illustrating a noise removing method according to an embodiment of the present invention.

Referring to FIG. 8, the noise reduction method according to the present embodiment includes steps that are processed in time series in the noise reduction system shown in FIG. 2. Therefore, even if omitted below, the above description of the noise canceling system illustrated in FIG. 2 is also applied to the noise canceling method according to the present embodiment.

In operation 810, the voice / noise section determination unit 21 receives an input signal having a noise signal and a voice signal.

In operation 815, the voice / noise section determination unit 21 determines a plurality of frames included in the input signal as a noise section or a voice section. Here, the noise section is a section including only the noise signal, not the voice signal, the speech section is a section including the noise signal and the voice signal.

In operation 820, the transition determiner 22 determines whether the current frame, that is, the nth frame F _n , is a voice interval. As a result of the determination, if the n th frame F _n is the voice interval, step 845 is performed, and if it is not the voice interval, step 825 is performed.

In operation 825, the transition determiner 22 determines whether the previous frame, that is, the (n−1) th frame F _{n −1} is a voice interval. As a result of the determination, when the (n-1) th frame (F _{n -1} ) is the voice interval, the transition from the voice interval to the noise interval is reported. In step 840, if it is not the voice interval, the transition is not reported. Follow step 830.

In operation 830, the noise estimating and updating unit 23 estimates noise in the current frame, that is, the nth frame.

In operation 835, the estimated noise storage unit 24 stores the estimated noise in the nth frame. At this time, it is determined based on the size or the total delay of each frame of the capacity of the estimated noise storage 24. For example, the estimated noise storage unit 24 may store the noises estimated in the frame of a.

In operation 840, the operation is performed again from the operation 820 on the next frame, that is, the (n + 1) th frame. In other words, it is determined whether the (n + 1) th frame is a voice section or a noise section, and whether or not a transition occurs in the (n + 1) th frame.

In operation 845, the transition determiner 22 determines whether the previous frame, that is, the (n−1) th frame F _n−1 is a noise period. As a result of the determination, when the (n-1) th frame (F _n-1 ) is a noise section, the transition from the noise section to the speech section is reported. In step 850, if the transition is not the noise section, the transition is not reported. Perform step 855.

In operation 850, the noise estimating and updating unit 23 loads the estimated noise in the (na) th frame F _{n -a} in the estimated noise storage unit 24, and loads the loaded noise into the nth frame (F _n). Update to the estimated noise used for the noise canceling operation. Thus, the estimated noise in at least one frame existing between the n th frame F _n and the (na) th frame F _{n -a} is not used in subsequent noise cancellation operation.

In operation 855, the noise removing unit 26 removes the noise in the n th frame F _n based on the updated estimated noise, that is, the noise estimated in the (na) th frame F _{n -a} .

In operation 860, the noise removing unit 26 outputs the speech signal from which the noise is removed from the input signal.

As described above, although the present invention has been described by way of limited embodiments and drawings, the present invention is not limited to the above-described embodiments, which can be variously modified and modified by those skilled in the art to which the present invention pertains. Modifications are possible. Accordingly, the spirit of the invention should be understood only by the claims set forth below, and all equivalent or equivalent modifications will fall within the scope of the invention.

In addition, the apparatus according to the present invention can be embodied as computer readable codes on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of the recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and also include a carrier wave (for example, transmission through the Internet). The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

10: voice / audio codec
20: noise reduction system
21: voice / noise section discriminating unit
22: transition judgment unit
23: noise estimation and update unit
24: estimated noise storage
25: converter
26: noise canceling unit

Claims (19)

Determining a plurality of frames included in an input signal having a noise signal and a voice signal as a noise section or a voice section;
Estimating noise in at least one noise frame determined as the noise section among the plurality of frames;
For a voice transition frame transitioned from the noise period to the voice interval among the plurality of frames, the noise estimated in the noise frame before the predetermined number of transition frames is regarded as an estimated noise for the voice transition frame. Removing noise from the speech transition frame using estimated noise; And
And removing the noise using the estimated noise in at least one voice frame determined as the voice interval among the plurality of frames. The method of claim 1,
And storing the noise estimated in the at least one noise frame in a buffer. The method of claim 2,
Removing noise from the voice transition frame,
Loading noise estimated in a noise frame before the voice transition frame into a predetermined number of noise frames;
Updating the loaded noise with the estimated noise for the speech transition frame; And
And outputting the speech signal by removing the estimated noise from an input signal corresponding to the speech transition frame. The method of claim 3,
The step of loading noise estimated from a noise frame before a predetermined number of the voice transition frame in the buffer may include loading noise estimated in a (na) th frame from the buffer when the voice transition frame is an nth frame. and,
The updating of the loaded noise with the estimated noise for the speech shifted frame may include updating the noise estimated in the (na) th frame with the estimated noise for the speech shifted frame, thereby generating the nth frame. Exclude noise estimated in at least one frame existing between and the (na) th frame in a subsequent noise cancellation step,
and n and a are natural numbers. The method of claim 4, wherein
A is determined based on at least one of a size of each frame and an overall delay period. The method of claim 4, wherein
And the capacity of the buffer is determined according to the a. The method of claim 1,
Estimating noise in the at least one noise frame,
Estimating noise in the at least one noise frame when the previous frame of the at least one noise frame is the noise period,
And if the previous frame of the at least one noise frame is the voice interval, noise is not estimated in the at least one noise frame. The method of claim 1,
And converting the input signal from time domain to frequency domain. The method of claim 8,
Removing noise from the at least one voice frame,
And removing the estimated noise from the converted input signal corresponding to the at least one voice frame to output the voice signal. delete A voice / noise section discriminating unit for discriminating a plurality of frames included in an input signal having a noise signal and a voice signal as a noise section or a voice section;
A transition determining unit determining whether a transition has occurred in at least one noise frame determined as the noise section or at least one voice frame determined as the speech section among the plurality of frames;
Estimating noise in the at least one noise frame, and for a voice transition frame in which a transition is made from the noise section to the speech section, noise estimated from a noise frame before the voice transition frame is a predetermined number. A noise estimating and updating unit updating the estimated noise with respect to the noise; And
And a noise canceller configured to remove noise from the speech transition frame or the speech frame by using the estimated noise. The method of claim 11,
And a noise estimator for storing noise estimated from the at least one noise frame. The method of claim 12,
The noise estimation and update unit,
The noise canceling system, characterized in that for loading the speech transition frame from the estimated noise storage unit, the noise estimated in a previous number of noise frames, and updating the loaded noise with the estimated noise for the speech transition frame . The method of claim 13,
The noise estimation and update unit,
If the speech shifted frame is an nth frame, the noise estimated in the (na) th frame is loaded from the estimated noise storage unit, and the noise estimated in the loaded (na) th frame is used for the speech shifted frame. By updating the estimated noise, the noise canceller is not provided to the noise canceller in at least one frame existing between the nth frame and the (na) th frame,
and n and a are natural numbers. The method of claim 14,
And a is determined based on at least one of the size of each frame and the delay period of the noise cancellation system. The method of claim 14,
The capacity of the estimated noise storage is determined according to the a. The method of claim 11,
The noise estimation and update unit,
Estimating noise in the at least one noise frame when the previous frame of the at least one noise frame is the noise period, and noise in the at least one noise frame when the previous frame of the at least one noise frame is the voice interval Noise canceling system, characterized in that does not estimate. The method of claim 11,
And a converter for converting the input signal from the time domain to the frequency domain. The method of claim 18,
The noise removing unit,
And outputting the speech signal by removing the estimated noise from the speech transition frame or the converted input signal corresponding to the speech frame.

Images