KR101591597B1 - Adaptive muting system and mehtod using g.722 codec packet loss concealment and steepest descent criterion - Google Patents

Abstract

An adaptive muting system and method using G.722 codec packet loss concealment and top-down method is disclosed. The adaptive muting system includes a decoding signal generator for normally decoding the encoded frame in the G.722 encoder to generate a decoded signal, a linear encoder for receiving the encoded frame, A loss hypothesis decoding signal generation unit for generating a loss hypothesis decoding signal by performing prediction based pitch repetition and a parameter determination unit for determining parameters of the sigmoid function using the decoding signal and the loss hypothesis decoding signal, Section.

Description

[0001] DESCRIPTION [0002] ADAPTIVE MUTING SYSTEM AND MEHTOD USING G.722 CODEC PACKET LOSS CONCEPT AND STEEPEST DESCENT CRITERION [

Embodiments of the present invention are directed to a technique for determining parameters of a sigmoid function using a best fit method in a packet loss concealment algorithm of a G.722 codec.

Generally, in a voice communication system using Voice over Internet Protocol (VoIP), packets are lost in the middle of packets. This packet loss reduces voice call quality. Accordingly, the conventional VoIP voice communication system compensates for packet loss using a packet loss concealment algorithm.

The G.722 codec is a broadband-based codec (CODEC) standardized by ITU-T and widely used for VoIP voice communications. For the G.722 codec, the parameters of the sigmoid function are determined using a grid search training technique.

Thus, in the case of using a training technique of the grid search method, the parameter of the once determined sigmoid function is fixed value which does not change afterwards, and is dependent on the training database.

Therefore, in G.722 Appendix IV standardized by the ITU-T that defines the packet loss concealment algorithm, when the parameter of the sigmoid function is determined using the training method of the grid search method, the discontinuity of the predetermined value It is difficult to achieve optimized muting by calculating the muting gain using an adaptive muting technique using curves.

In other words, the parameters of the sigmoid function are dependent on the training data, and it is difficult to cope with a short time variation of the voice since the determined parameter is used constantly and unchanged.

Therefore, there is a need for a technique capable of updating the parameters of the sigmoid function on a sample-by-sample basis according to the characteristics of each speech without setting the parameters of the sigmoid function fixed.

In order to solve the problem of using the parameters of the fixed sigmoid function, the present invention determines and updates the parameter of the sigmoid function on a sample basis using the best-fit method to improve the voice quality .

The adaptive muting system according to an embodiment of the present invention includes a decoding signal generation unit for decoding a G.722 encoder normal decoding / G.722 encoding to generate a decoded signal, A loss hypothesis decoding signal generation unit for performing a linear prediction based pitch repetition on a previous frame that has been received by the loss prediction unit and has no packet loss to generate a loss hypothesis decoding signal, And a parameter determination unit for determining a parameter of the sigmoid function using the decoded signal.

According to an aspect, the parameter of the sigmoid function may be determined and updated every frame using a Steepest Descent Criterion.

According to another aspect of the present invention, there is provided a decoding apparatus for decoding a lossy encoded frame, the apparatus comprising: a determination unit for determining whether the encoded frame is a lost frame; And a parameter updating unit for updating the parameter of the sigmoid function corresponding to the current frame with the parameters of the sigmoid function corresponding to the previous frame as determined to be the lost frame, .

According to another aspect of the present invention, the encoded frame is a frame in which packet loss has not occurred, and the loss hypothesis decoding signal generation unit performs the linear prediction based pitch repetition on the assumption that a packet loss occurs in the encoded frame .

According to another aspect of the present invention, the parameter determination unit may determine the parameter of the sigmoid function so that an error signal indicating a difference between the reconstructed signal reconstructed from the encoded frame and the decoded signal is reduced or minimized.

According to another aspect of the present invention, there is provided a signal processing apparatus comprising: a muting gain calculator for calculating a muting gain using parameters of the sigmoid function; and a restoring unit for restoring a current frame using the muting gain and the lost- And a generating unit.

According to another aspect of the present invention, the decoding signal generator may decode a signal corresponding to a low band of a lower band and a higher band signal included in the encoded frame. The loss hypothesis decoding signal generation unit may perform the linear prediction based pitch repetition on the signal corresponding to the low band.

According to another aspect of the present invention, the loss hypothesis decoding signal generation unit generates the loss hypothesis decoding signal based on the linear prediction-based pitch repetition signal for a signal corresponding to the low band among a lower band and a higher band signal included in the encoded frame, Can be performed.

An adaptive muting method according to an embodiment of the present invention includes the steps of normal decoding / G.722 decoding a frame encoded in a G.722 encoder to generate a decoded signal, Generating a lossy hypothesis decoding signal by performing linear prediction based pitch repetition on a previous frame in which no packet loss has occurred, and generating a loss hypothesis decoding signal by using the lossy hypothesis decoding signal, And determining a parameter of the function.

According to an aspect, the parameter of the sigmoid function may be determined and updated every frame using a Steepest Descent Criterion.

According to another aspect of the present invention, there is provided a method of decoding a lost frame, the method comprising: determining whether the encoded frame is a lost frame; determining whether the encoded frame is a lost frame; determining a parameter of a sigmoid function corresponding to a current frame, Updating the parameter of the sigmoid function corresponding to the current frame with the parameter of the sigmoid function corresponding to the previous frame according to the determination as the lost frame, .

According to another aspect, the encoded frame is a frame in which no packet loss has occurred, and the step of generating the lossy hypothesis decoding signal includes: Can be performed.

According to another aspect of the present invention, the determining of the parameter of the sigmoid function may include determining a parameter of the sigmoid function so that an error signal indicating a difference between a reconstructed signal obtained by reconstructing the encoded frame and the decoded signal is reduced or minimized, Can be determined.

According to another aspect, the method further comprises calculating a muting gain using the parameters of the sigmoid function, and restoring the current frame using the muting gain and the loss hypothesis decoding signal .

According to another aspect of the present invention, the step of generating the loss hypothesis decoding signal may include a step of, for a signal corresponding to the low band among a lower band and a higher band signal included in the encoded frame, Based pitch iteration.

According to another aspect of the present invention, the step of generating the decoding signal may include decoding a signal corresponding to a low band of a lower band and a higher band signal included in the encoded frame, The generating of the signal may perform the linear prediction based pitch iteration on the signal corresponding to the low band.

According to the present invention, the voice quality can be improved by determining and updating the parameters of the sigmoid function on a sample basis using the best-fit method.

1 is a block diagram illustrating an overall configuration of a G.722 decoder according to an embodiment of the present invention.
2 is a block diagram illustrating a configuration of an adaptive muting system according to an embodiment of the present invention.
3 is a block diagram illustrating a process of generating a restoration signal in the LP-based pitch repetition unit of FIG.
FIG. 4 is a flowchart illustrating an operation of generating a restoration signal using a packet loss concealment method and a top-down method according to an embodiment of the present invention.
5 is a diagram illustrating a waveform of a restoration signal generated using a muting gain according to an embodiment of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an overall configuration of a G.722 decoder according to an embodiment of the present invention.

Referring to FIG. 1, the G.722 decoder 100 includes a low-band ADPCM (Low-Band Adaptive Differential Pulse Code Modulation) decoder 101, a low-band ADPCM state update unit 102, Based pitch repetition unit 103, crossfading 104, a high-band ADPCM decoder 105, a high-band ADPCM state differential update unit 105, A Pulse Code Modulation State Update 106, a Pitch Repeat 107, a A 107, a Hpost 109, a B 110, and a Quadrature Mirror Filter Synthesis Filter Bank 111, . ≪ / RTI >

1, the G.722 decoder 100 includes a lower band ADPCM State Update (102), a LP-based pitch An LP-based pitch repetition unit 103, a crossovering unit 104, a high-band ADPCM state update unit 106, a pitch repetition unit 107, an Hpost 109 ) And B (110).

In Fig. 1, Zl (n) may be the previous frame of the nth sample where no packet loss occurred. Here, the previous frame may be a frame received immediately before the current frame among the encoded frames continuously received encoded in the G.722 encoder. For example, if frame 1, frame 2, and frame 3 encoded in the G.722 encoder are successively received and the current frame is frame 2, the previous frame may be frame 1.

For example, in FIG. 1, when packet loss does not occur in the previous frame but packet loss occurs in the current frame, the G.722 decoder 100 uses the previous frame Zl (n) To generate the reconstruction signal yl (n) of the n-th sample. In other words, when a bad frame in which a packet loss occurs is received in the current frame, the G.722 decoder 100 can recover the current frame using the previous frame Zl (n) of the nth sample.

When the next frame xl (n) of the n-th sample is a frame in which no packet loss has occurred after restoring the current frame, the cross fading 104 is performed in such a manner that the reconstruction signal yl (n) (N) of the nth sample and the next frame xl (n) of the nth sample to reduce or eliminate the discontinuity between the next frame xl (n) of the nth sample.

Here, the next frame may be a frame received after the current frame among the frames continuously received and encoded in the G.722 encoder, and may correspond to a good frame in which packet loss has not occurred. For example, if Frame 1, Frame 2, and Frame 3 encoded in the G.722 encoder are successively received and the current frame is Frame 2, the next frame may be Frame 2.

The G.722 codec may be a sub-band codec configured as a high band and a low band. At this time, the voice signal may exist in the low band. Then, the LP-based pitch iteration performing unit 103 may restore the voice signal existing in the low band using a linear prediction based pitch repetition algorithm. At this time, the G.722 decoder 100 can reduce the calculation complexity by restoring the signal existing in the high band using only the pitch period of the previous frame Zl (n) of the n-th sample.

2 is a block diagram illustrating a configuration of an adaptive muting system according to an embodiment of the present invention.

2, the adaptive muting system 200 includes a determination unit 201, a decoding signal generation unit 202, a loss hypothesis decoding signal generation unit 203, a muting processing unit 204, a buffer 208, Unit 207, and a restoration signal generator 209. [0050]

First, the determination unit 201 receives the encoded frame from the G.722 encoder, and can determine whether the encoded frame is a frame in which packet loss has occurred. Here, the G.722 encoder is an encoder standardized by ITU-T, and there is no change with the standard, so a detailed description will be omitted. In addition, G.722 is a sub-band codec, and most voice signals may exist in a lower band.

At this time, if the encoded frame is a frame in which packet loss has not occurred, the decoding signal generation unit 202 can generate an n-th sample decoded signal dl (n) as normal decoding of the encoded frame have. For example, the decoding signal generation unit 202 may perform G.722 decoding on an encoded frame in which packet loss has not occurred. Accordingly, the decoded signal dl (n) of the n-th sample is a clear signal and can be a desired signal.

The lossy decoding signal generation unit 203 may generate a lossy decoding signal by performing LP based Pitch Repetition on a previous frame received before the encoded frame and having no packet loss have.

For example, the lossy decoding signal generation unit 203 may assume that a loss occurs in the encoded frame. The loss hypothesis decoding signal generation unit 203 performs LP based pitch repetition on the previous frame zl (n) of the encoded frame, and outputs the loss hypothesis decoding signal ylpre (n) (n) < / RTI >

The muting processing unit 204 may include a parameter determination unit 205 and a muting gain calculation unit 206 for determining the parameters of the sigmoid function using the best-fit method.

The parameter determination unit 205 can determine the parameter of the sigmoid function by applying the most probable descent method to the decoding signal dl (n) of the n-th sample and the loss hypothesis decoding signal ylpre (n). At this time, the parameter determination unit 205 can determine the parameter of the sigmoid function every frame.

For example, the parameter determining unit 205 may determine the difference between the reconstructed signal yl (n) of the n-th sample and the decoded signal dl (n) of the n-th sample obtained by restoring the encoded frame based on Equations (1) The parameters of the sigmoid function can be determined such that the error signal e (n) of the nth sample representing the difference is reduced or minimized.

(N) is the muting gain of the n-th sample, ylpre (n) is the loss of the n-th sample, y (n) Decoded signal.

According to Equation (1), the reconstruction signal yl (n) of the nth sample can be calculated by multiplying the muting gain G (n) of the nth sample by the loss hypothesis decoding signal ylpre (n) of the nth sample. At this time, the loss hypothesis decode signal ylpre (n) of the nth sample can be obtained from the loss hypothesis decoding signal generation section 203, but the restoration signal yl (n) of the nth sample and the muting gain G I do not know yet. Accordingly, the parameter determination unit 205 can determine the parameters of the sigmoid function based on the following Equation (2) indicating the relationship between the muting gain and the parameters of the sigmoid function.

In Equation 2, α and β is a parameter that determines the shape of the sigmoid function, n ₀ may be offset.

For example, when packet loss occurs continuously for four or more frames, the muting gain G (n) of the n-th sample may be set to zero. Using Equation (2), the muting curve is continuous for the frame and can obtain a flexible muting gain.

At this time, the best-fit method is an algorithm that minimizes the difference between the reconstructed signals yl (n) of the n-th sample in a desired signal. Accordingly, the adaptive muting system 200 can generate the restored signal by performing the LP based pitch repetition on the good frame in which the packet loss does not occur in order to utilize the fastest descent method. In other words, the parameter determination unit 205 multiplies the decoded signal dl (n) of the n-th sample and the loss-based decoding signal ylpre (n) of the n-th sample, which are the results of normal decoding based on Equation .

In Equation (3), e (n) is the error signal of the nth sample, dl (n) is the decoded signal of the nth sample, and yl (n) is the restored signal of the nth sample. At this time, according to Equation 1, the reconstruction signal yl (n) of the nth sample can be expressed by the muting gain G (n) of the nth sample and the loss hypothesis decoding signal ylpre (n) of the nth sample. Then, the parameter determination unit 205 can substitute the equation (1) into the equation (3). Then, according to equation (2), the muting gain G (n) of the n-th sample may be represented by parameters α, β, and n ₀ for determining the gradient of the sigmoid function. Accordingly, the parameter determination unit 205 can substitute the equation (2) into the equation (3). Then, the error signal e (n) of the n-th sample can be expressed by subtracting the product of the parameter of the sigmoid function and the loss assumption signal in the decoded signal dl (n) of the n-th sample.

Based on Equation (3), the parameter determination unit 205 determines whether the error signal indicating the difference between the n-th sample decoded signal dl (n) and the n-th sample lossy decoding signal ylpre (n) of it may determine the parameters α, β, and n _0. At this time, the error signal can be expressed as the following equation (4) as a cost function.

In Equation (4), the cost function J (?,?) May be a function determined by?,?. Then, the parameter determination unit 205 can determine α and β in the direction of minimizing the cost function J (α, β) by using the most rapid method.

Then, the buffer 208 can temporarily store the parameters of the sigmoid function determined for each frame in the parameter determination unit 205. [ For example, the buffer 208 may temporarily store a parameter of a sigmoid function corresponding to frame 1, a parameter of a sigmoid function corresponding to frame 2, and a parameter of a sigmoid function corresponding to frame 3.

Then, the parameter update unit 207 may update the parameter of the sigmoid function determined for each frame.

For example, when the determination unit 201 determines that the encoded frame is not a lost frame, the parameter update unit 207 updates the parameter of the sigmoid function corresponding to the current frame to the sigmoid function determined using the loss- You can update it with the parameters of the de function. In other words, the parameter update unit 207 can update the parameter determined based on Equation (4) above with parameters of the sigmoid function corresponding to the current frame.

If it is determined that the frame is a lost frame, the parameter update unit 207 refers to the buffer 208 to update the parameter of the sigmoid function corresponding to the current frame with the parameter of the sigmoid function corresponding to the previous frame have.

For example, the parameter update unit 207 may update the parameters of the sigmoid function based on the following equations (5) to (8).

는 아래의 수학식 6과 같이 표현될 수 있다.In the equation (5),? (N) is a parameter for determining the shape of the sigmoid function in the nth sample,? (N + 1) is a parameter for determining the shape of the sigmoid function in the ego,

Can be expressed by Equation (6) below.

Equation (6) can represent a value obtained by differentiating G with respect to?.

As described above, the parameter update unit 207 can update the parameter a of the sigmoid function every frame based on Equations (5) and (6).

The parameter update unit 207 may update the parameter beta of the sigmoid function every frame based on Equations (7) and (8) below.

는 아래의 수학식 8과 같이 표현될 수 있다.In Equation (7),? (N) is a parameter for determining the shape of the sigmoid function in the nth sample,? (N + 1) is a parameter for determining the shape of the sigmoid function in the ego,

Can be expressed as Equation (8) below.

Equation (8) can represent a value obtained by differentiating G with respect to?.

As described in Equations (1) to (8) above, the parameters of the sigmoid function can be determined every frame so as to minimize the cost function using the best-fit method. Then, the parameter update unit 207 can continuously update the parameters of the sigmoid function with the parameters determined every frame to minimize the cost function. The parameters of the sigmoid function obtained in a frame in which no packet loss occurs can be used for adaptive muting in a frame in which packet loss occurs.

범위 내에서 결정될 수 있다.Considering the shape of the sigmoid function, n ₀ can be set to 150, and α and β

Can be determined within a range.

As described above, when the parameter of the sigmoid function is determined and updated, the muting gain calculation unit 206 can calculate the muting gain based on Equation (2) above. Then, the restoration signal generator 209 can generate a restoration signal based on Equation (1). For example, the reconstruction signal generator 209 multiplies the restoration signal yl (n) in the n-th sample by the product of the muting gain G (n) in the n-th sample and the loss hypothesis decoding signal ylpre Can be generated.

At this time, as the parameter of the sigmoid function is continuously updated, the muting gain calculation unit 206 can calculate the muting gain G (n) in the n-th sample using the parameters of the updated sigmoid function. In other words, the muting gain G (n) in the nth sample may have an adaptive value. Accordingly, the reconstruction signal generation unit 209 gradually multiplies the adaptive muting gain G (n) in the n-th sample by the loss hypothesis decoding signal ylpre (n) in the n-th sample to gradually attenuate the signal, Lt; / RTI > Accordingly, when a restored signal is generated using the adaptive muting gain rather than using the lost presumed decoded signal ylpre (n) in the nth sample at the time of packet loss, unwanted noise such as a mechanical noise can be reduced or eliminated have.

3 is a block diagram illustrating a process of generating a restoration signal in the LP-based pitch repetition unit of FIG.

FIG. 3 is a block diagram illustrating a process of generating a restoration signal by applying a packet loss concealment method and a method of optimizing the strength of a packet according to an exemplary embodiment of the present invention. Referring to FIG.

Referring to FIG. 3, a reconstructed signal yl (n) in the nth sample corresponding to the current frame can be generated using the previous frame Zl (n) in the nth sample in which no packet loss has occurred. At this time, the LP analysis (LP analysis) 301 can extract the LPC coefficients by performing the linear prediction analysis on the previous frame Zl (n) in the nth sample. The LTP analysis 302 extracts the pitch interval T ₀ of the previous frame Zl (n) in the n-th sample and the signal classification 303 extracts the previous frame Zl (n) n) can be extracted.

Then, the modification and pitch repetition unit 304 can generate the lost presumed decoded signal ylpre (n) in the n-th sample using the LPC coefficient, the class, and the pitch interval T ₀ . Here, LP analysis (LP analysis) 301, LTP analysis (LTP analysis) 302, signal classification (303), and correction and pitch repetition (304) And may be included in the signal generation unit.

The muting factor computation unit 305 determines and updates the parameter of the sigmoid function for each frame and adaptively calculates the muting gain using the updated parameter. Here, the muting factor calculator 305 may correspond to the muting processor 204 of FIG.

Then, the reconstruction signal generator 306 may generate the reconstruction signal yl (n) at the n-th sample by multiplying the adaptive muting gain by the lost presumption decoded signal ylpre (n) at the n-th sample. The restoration signal generator 306 may generate a restoration signal by reducing or eliminating unnecessary noise caused by periodic repetition of the pitch by using the adaptive muting gain.

FIG. 4 is a flowchart illustrating an operation of generating a restoration signal using a packet loss concealment method and a top-down method according to an embodiment of the present invention.

In Fig. 4, the operation of generating the restoration signal can be performed by the adaptive muting system of Fig.

First, in step 401, the adaptive muting system 200 may receive an encoded frame from a G.722 encoder.

Then, in step 402, the adaptive muting system 200 may determine whether the encoded frame is a lost frame in which packet loss occurred.

If it is determined that the frame is not a lost frame (No in step 402), in step 403, the adaptive muting system 200 performs normal decoding on the encoded frame to generate a decoded signal dl (n) < / RTI > For example, the adaptive muting system 200 may perform G.722 decoding. Then, the decoded signal dl (n) in the nth sample may be a desired signal with a clean signal.

Next, in step 404, the adaptive muting system 200 performs LP based pitch repetition, assuming that a packet loss occurs in a frame in which packet loss has not occurred in order to apply the best luck method . For example, the adaptive muting system 200 may perform LP based pitch repetition on the previous frame Zl (n) in the nth sample as assuming packet loss, It is possible to generate the hypothesis decoding signal ylpre (n).

Then, in step 405, the adaptive muting system 200 determines a parameter of the sigmoid function using the lost presumption decoded signal ylpre (n) in the nth sample and the decoded signal dl (n) in the nth sample .

For example, the adaptive muting system 200 may determine the parameters of the sigmoid function such that the error signal indicative of the difference between the decoded signal and the reconstructed signal is minimized or decreased based on Equations 1 to 5 above. At this time, the adaptive muting system 200 may determine the parameters of the sigmoid function for each frame.

Then, in step 406, the adaptive muting system 200 may update the parameters of the determined sigmoid function for each frame.

Then, in step 407, the adaptive muting system 200 may calculate the muting gain using the parameters of the updated sigmoid function. At this time, as the parameters of the sigmoid function are updated every frame, the adaptive muting system 200 may adaptively calculate the muting gain G (n) in the nth sample.

Then, in step 408, the adaptive muting system 200 can generate a restoration signal using the adaptive muting gain G (n) in the n-th sample and the loss hypothesis decoding signal ylpre (n) in the n-th sample have.

For example, the adaptive muting system 200 multiplies the adaptive muting gain G (n) at the n-th sample by the loss hypothesis decoding signal ylpre (n) at the n-th sample, yl (n) < / RTI >

Thus, the adaptive muting system 200 can adaptively calculate the muting gain by continuously updating the parameters of the sigmoid function, assuming that packet loss has occurred even if no packet loss has occurred.

At this time, if a packet loss occurs, that is, if the encoded frame is determined to be a lost frame (402: YES), the adaptive muting system 200 performs muting using the parameters of the sigmoid function corresponding to the previous frame The gain can be calculated. Here, the previous frame may correspond to a frame received before the lost frame, and no packet loss has occurred.

For example, the adaptive muting system 200 refers to a buffer to determine a parameter of a sigmoid function corresponding to a previous frame in which packet loss has not occurred as a parameter of a sigmoid function corresponding to a current frame in which a packet loss occurs And update the parameters. In step 409, the adaptive muting system 200 may calculate the muting gain using parameters of the sigmoid function corresponding to the updated current frame.

In this way, when the packet loss occurs, the adaptive muting system 200 can adaptively calculate the muting gain by using the parameters of the sigmoid function corresponding to the previous frame in which packet loss has not occurred. Then, the adaptive muting system 200 generates a restored signal by using the calculated adaptive muting gain and the lost-hypothesized decoded signal, and then generates a restored signal by removing unnecessary noise such as a mechanical sound.

5 is a diagram illustrating a waveform of a restoration signal generated using a muting gain according to an embodiment of the present invention.

5, a voice file of the NTT database is used for verifying the performance, 501 is a waveform when a restoration signal is generated without muting, 502 is a waveform when a restoration signal is generated according to a muting method defined in the standard And 503 is a waveform when a restoration signal is generated using the adaptive muting gain proposed in the present invention.

FIG. 5 also includes waveforms when the muting does not operate in order to confirm the importance of muting (501). In FIG. 5, a period from 0 to 0.01 seconds is a period in which no packet loss occurs, and a period from 0.01 to 0.05 seconds corresponds to a period in which a packet loss occurs. At this time, when the restored signal yl (n) in the nth sample and the desired signal dl (n) in the nth sample are compared in 501 to 503, the restored signal 503, which generates the restored signal using the muting gain according to the present invention, It can be seen that the signal is restored most like the desired signal. Here, the desired signal may be a decoded signal.

Also, referring to FIG. 5, when muting is not performed, it is confirmed that the same waveform is repeated at the same pitch period, and unnecessary noise and mechanical noise are generated. Referring to 502, it can be seen that the muting method according to the standard has more muting than the adaptive muting method according to the embodiment of the present invention.

Table 1 below shows the results for segmental SNR, PESQ, and _Cov1 , which are objective voice quality metrics for various packet loss rates, and the results for MOS, a subjective voice quality evaluation index. For the MOS experiment, a total of 10 participants rated each voice subjectively from 1 to 5 points.

Packet loss rate Quality evaluation index Comparison algorithm Algorithm for muting according to existing standard The adaptive muting method according to the present invention 3% Seg. SNR 24.37 24.42 PESQ 3.40 3.42 C _ovl 3.69 3.80 MOS 3.39 3.45 6% Seg. SNR 21.08 21.14 PESQ 3.01 3.13 C _ovl 3.34 3.57 MOS 3.13 3.19 10% Seg. SNR 17.73 17.83 PESQ 2.76 2.92 C _ovl 2.86 3.21 MOS 2.64 2.80 15% Seg. SNR 13.21 13.36 PESQ 2.32 2.55 C _ovl 2.41 2.76 MOS 2.16 2.36

According to Table 1, it can be seen that the adaptive muting method according to an embodiment of the present invention performs muting more effectively than the muting method defined in the standard. This may indicate that the parameters of the sigmoid function determined using the best fit method are optimized for muting. As described above, by performing the muting using the parameter of the sigmoid function optimized for muting to generate the restored signal, the noise such as the mechanical noise can be removed or reduced and the voice quality can be improved.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (8)

A decoding signal generator for normally decoding the encoded frames in the G.722 encoder to generate a decoded signal;
A loss hypothesis decoding signal generation unit for generating a loss hypothesis decoding signal by performing linear prediction based pitch repetition on a previous frame which is a frame in which packet loss has not occurred among the encoded frames;
A parameter determination unit for determining a parameter of the sigmoid function using the decoded signal and the loss hypothesis decoding signal,
A muting gain calculator for calculating a muting gain using the parameter of the sigmoid function; And
A restoration signal generation unit for restoring a current frame by using the muting gain and the lost-
Lt; / RTI >
The loss hypothesis decoding signal generation unit may include:
Extracts an LPC coefficient, a pitch interval (T ₀ ) of a previous frame and a class of a previous frame with respect to the previous frame, generates the loss hypothesis decoding signal based on the extracted LPC coefficient, pitch interval, and class ,
Wherein the previous frame indicates a frame received before the current frame of the encoded frames
Wherein the adaptive muting system comprises: The method according to claim 1,
The parameters of the sigmoid function are:
(Steepest Descent Criterion) is used to determine and update each frame. The method according to claim 1,
A determination unit for determining whether the encoded frames are lost frames; And
Updating a parameter of a sigmoid function corresponding to a current frame with a parameterer of a sigmoid function determined using the loss hypothesis decoding signal as it is determined that the frame is not the lost frame, A parameter update unit for updating the parameter of the sigmoid function corresponding to the current frame with the parameter of the sigmoid function corresponding to the previous frame,
≪ / RTI > further comprising an adaptive muting system. The method according to claim 1,
The encoded frames are frames in which packet loss has not occurred,
The loss hypothesis decoding signal generation unit may include:
And performs the linear prediction based pitch iteration on the assumption that a packet loss has occurred in the encoded frames. The method according to claim 1,
Wherein the parameter determination unit comprises:
Wherein the parameter of the sigmoid function is determined such that an error signal indicating a difference between the reconstructed signal reconstructed from the encoded frames and the decoded signal is reduced or minimized. delete The method according to claim 1,
The loss hypothesis decoding signal generation unit may include:
Wherein the linear prediction based pitch repetition is performed on a signal corresponding to the low band among low and high band signals included in the encoded frames. Normal decoding the encoded frames in the G.722 encoder to generate a decoded signal;
Generating a loss hypothesized decoding signal by performing linear prediction based pitch repetition on a previous frame, which is a frame in which no packet loss occurred, from the encoded frames;
Determining a parameter of the sigmoid function using the decoded signal and the lost hypothesized decoded signal;
Calculating a muting gain using parameters of the sigmoid function; And
And restoring a current frame using the muting gain and the lossy decoding signal
Lt; / RTI >
Wherein the step of generating the lost hypothesized decoded signal comprises:
Extracts an LPC coefficient, a pitch interval (T ₀ ) of a previous frame and a class of a previous frame with respect to the previous frame, generates the loss hypothesis decoding signal based on the extracted LPC coefficient, pitch interval, and class ,
Wherein the previous frame indicates a frame received before the current frame of the encoded frames
Wherein the adaptive muting method comprises:

Images