KR20070059860A - Method and apparatus for restoring digital audio packet loss - Google Patents

Abstract

A method and an apparatus for recovering digital audio packet loss are provided to reduce audio quality deterioration by hiding a lost audio frame and improve accuracy of recovering an audio frame loss. A method for recovering digital audio packet loss includes a step of detecting whether an audio frame included in a received audio packet is lost, a step of detecting whether continuous audio frames are lost when an audio frame included in the received audio packet is lost, and a step of detecting a pitch from frames before lost audio frames when the continuous audio frames are lost and repeating the detected pitch to replace the lost audio frames.

Description

Method and apparatus for restoring digital audio packet loss

1 is a block diagram of a lost frame recovery apparatus according to the present invention;

2 is a structural diagram showing an example of an audio bitstream frame;

3 is a flowchart illustrating a procedure of a loss concealment method according to an embodiment of the present invention;

4 is a flowchart illustrating a first loss restoration method occurring in the first restoration module;

5 is a diagram for explaining a pitch search step;

FIG. 6 is a diagram illustrating a method for performing OLA at a connection portion of a frame and a pitch before loss; FIG.

7 is a flowchart illustrating a second loss restoration method occurring in the second restoration module;

8 shows the step of applying an OLA at the concatenation of the pitch obtained from the pre-loss buffer and the pitch obtained from the post-loss buffer.

<Brief description of the main parts of the drawing>

100: audio bitstream 110: lost frame detector

120: decoder 130: frame buffer before loss

140: frame buffer after loss 150: lost frame restoration module

152: first restoration module 154: second restoration module

160: audio signal

The present invention relates to a method and apparatus for recovering a lost audio signal at a receiving end of a digital multimedia broadcasting (DMB) device. When a lost frame is received, the lost frame is recovered using a frame before loss and a frame after loss. The present invention relates to a method and an apparatus for recovering.

In a broadcast network such as a DMB, data in a packet form is transmitted and received, and an audio frame is stored in the data area of the packet. In the case of the DMB network, packet loss is not severe. However, in the case of a DMB receiver that needs to receive and reproduce multimedia in real time, packet loss occurs due to transmission problems, packet processing problems, and synchronization with video data. . In a broadcast receiver requiring high sound quality, such packet loss leads to a loss of audio included therein, thereby degrading sound quality and causing a synchronization problem with a video. Therefore, lost concealment and loss concealment techniques are needed to enable real-time audio services.

To implement the real-time audio service, a receiver-based error concealment method that does not incur additional delay is suitable, but since it does not have any information on lost frames, the loss rate is low (less than 15%) and short time (4 ~ 40ms) can be applied at interval loss. If you use the information before the loss for too long to compensate for the loss, the audio is unnatural because it is restored by artificial sound. Due to this characteristic, most error concealment algorithms apply attenuation of the signal and replace it with silence in case of more than 60ms.

On the other hand, when terrestrial DMB service is active, an error concealment algorithm suitable for a standard audio decoder is required, and in a broadcasting system requiring high sound quality, the lower part of the loss section is compensated with frame information after loss, thereby reducing the sound quality degradation due to lost frames. Needs to be.

However, the MPEG-4 Bit Sliced Arithmetic Coding (BSAC) audio decoder, which is the standard of the audio compression technology of the domestic terrestrial DMB, does not have an algorithm for concealing the audio frame when it is lost. This is a factor that can cause various problems when the frame loss occurs, it is a factor that degrades the sound quality.

On the other hand, looking at the prior patent (recovery of the Republic of Korea Patent No. 341391, "Adaptive additional transmission method and packet loss recovery method for the interactive audio service and the audio input and output control device of the multimedia computer for this") regarding the recovery of the packet loss, The patent includes an additional transmission processor composed of a transmitting side and a receiving side, and the transmitting side dynamically sets the additional transmission scheme using the loss information fed back from the receiving side, and the receiving side recovers the lost packet using the additional data. I'm using the method. In the above method, since the loss information needs to be fed back from the receiver in order to configure the additional information, there is a problem in that a delay caused by the feedback is inevitably generated, and the degree of delay of the additional information is changed according to the loss rate, thereby increasing the loss rate. As a result, the decoding delay time is increased and bandwidth is increased by additional information.

In addition, referring to other prior patents related to the recovery of packet loss (Korean Patent No. 238324, "Error concealment method and apparatus for audio signals"), the prior patent consists of a triangular converter, a selector and an error concealment data generator, The energy concentration is used to find the transform coefficients that best characterize the audio signal components of the frames before and after the lost frame and restore the audio data. However, in the above patent, a system is complicated by applying a triangular converter for conversion into a frequency domain, and there is a disadvantage in that it is not easy to implement, and therefore, it is not appropriate to be applied to a DMB system in which continuous loss does not occur significantly.

In order to solve the above-described problem, an object of the present invention is to provide a method and apparatus for recovering a lost frame by using a pre-loss frame and a post-loss frame when a lost frame is received.

In order to achieve the above object, the audio packet loss recovery method of the present invention includes detecting whether an audio frame stored in a received audio packet is lost, and detecting a loss of a continuous audio frame when the audio frame is lost. And a loss reconstruction step of retrieving a pitch from pre-loss frames and replacing the lost frame by repeating the retrieved pitch when the audio frame is continuously lost.

In addition, the audio packet loss recovery method of the present invention comprises the steps of detecting the loss of the audio frame stored in the received audio packet, and if the loss of the audio frame occurs, detecting whether the loss of the continuous audio frame, and And a loss retrieval step of retrieving a pitch frame from pre-loss frames and post-loss frames if a lossless frame is input immediately after a loss of the audio frame, and repeating the retrieved pitch frames to replace the lost frame. It is done.

In addition, the apparatus for recovering an audio packet loss of the present invention includes a lost frame detector for detecting whether an audio frame stored in a received audio packet is lost, a decoder for decoding the received audio frame, and temporarily storing frames before the lost frame. A pre-loss frame buffer, a post-loss frame buffer for temporarily storing frames after the lost frame, a first reconstruction module for replacing a pitch retrieved from the pre-loss frame buffer with a lost frame, and the pre-loss frame buffer. And a second reconstruction module that combines the retrieved pitch and the retrieved pitch from the lost frame buffer and replaces the lost frame with a lost frame.

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

1 is a block diagram of an apparatus for recovering a lost frame according to an embodiment of the present invention.

Referring to FIG. 1, the restoration apparatus checks whether a packet is lost with respect to the received audio bitstream packet 100, and detects whether a frame is lost or not, and the received audio bitstream packet. A decoder 120 that decodes the signal, a pre- and post-loss frame buffer 130 and 140 for temporarily storing frames that have passed through the decoder, and restores a lost frame from a pre- / post-loss frame when a lost frame occurs. The lost frame recovery module 150 includes the first and second recovery modules 152 and 154.

The audio bitstream packet 100 is received at a receiving end (not shown) as data in a form in which packet data are arranged in a line and transmitted. The audio bitstream packet 100 is composed of a header section and a data section. The header section includes information on a source, a destination, an identifier of a packet, a header checksum, and the like. In the data area, an audio frame to be transmitted is stored. In the embodiment of the present invention, since 44.1 Khz is used as the sampling frequency, and one frame is composed of 1024 samples, the playback time of one frame is approximately 23 ms (1024/43100).

Since the loss of the packet means the loss of the frame, the loss frame detector 110 determines whether the lost frame is lost by determining whether the received packet is lost. On the other hand, the bitstream represents a flow of data continuously transmitted through the serial communication line one packet at a time. Therefore, in the present invention, which is performed afterwards, it is determined whether or not there is a loss for each packet transmitted continuously and then decoded.

The lost frame detector 110 receives the audio bitstream packet 100 and determines whether the received packet is lost by using a header checksum stored in the packet, and if the received packet is lost, the frame stored in the packet is also lost. It is determined that the loss will be described later will be described. If the packet is lost according to the determination result, information about the packet loss is transmitted to the lost frame recovery module 150, and information such as the number of lost frames and the identifier of the lost frame is also transmitted. do.

Meanwhile, the decoder 120 decodes the frames received from the lost frame detector 110 into an audio signal. When transmitting a digital signal, the transmitting side converts the input data into an arbitrary code sequence to prevent the timing information from being lost in the continuous data and to prevent the mixing / modulation phenomenon occurring during the transmission process. A decoding process is performed to recover original data from a random code sequence.

When the decoding process is performed, the frames stored in the data region of the audio packet are transmitted to the frame buffers 130 and 140 before and after the loss.

The before and after loss frame buffers 130 and 140 temporarily store the decoded signals. In this case, if no packet loss occurs, that is, if there is no frame loss, the packet buffer is temporarily stored in the frame buffer 130 before the loss, and is output as the audio signal 160 after a predetermined time. However, if a lost packet occurs, frames of the packet transmitted after the packet loss occurs, that is, frames after the loss, are temporarily stored in the lost frame buffer 140.

 The lost frame recovery module 150 receives information on packet loss from the lost frame detector 110 to recover lost frames. The reconstruction module 150 includes a first reconstruction module 152 for reconstructing a lost audio signal from a pre-loss audio signal and a second reconstruction module 154 for reconstructing a lost audio signal from a post-loss audio signal and a pre-loss audio signal. ). Reference will be made to the example of the audio bitstream frame shown in FIG. 3 to look at the operation of the first and second reconstruction modules.

2 is a structural diagram illustrating an example of the audio bitstream frame, in which frames corresponding to data of each packet are arranged in a line. Darkly shaded A _n , A _{n + 1} frames correspond to the lost frames.

Referring to FIG. 2, the operation in the restoration module will be briefly described. In order to restore the A _n frame among the lost A _n and A _{n + 1} frames, since A _{n + 1} frames are lost, the information after the loss cannot be used. A frame is recovered using A _n-3 , A _n-2 and A _n-1 frames, which are processed in the first reconstruction module. However, since the lost frame can be used to recover the A _{n + 1} frame, the lost frame is restored using A _n-2 and A _n-1 frames before the loss and A _{n + 2} frames after the loss. 2 is processed in the restoration module.

That is, when a lost frame occurs, it is determined whether the frame has been lost in succession, and if the lost frames are continuously made, the first reconstruction module restores using only the frames before the loss since the frames after the loss cannot be used. If not, the second recovery module 154 recovers using the pre-lost frame and the post-lost frame. Before describing the detailed operation of the restoration modules, the overall flow of the loss concealment method according to the present invention will be described.

3 is a flowchart illustrating a procedure of a loss concealment method according to an embodiment of the present invention.

Referring to FIG. 3, first, a packet is received (310), and a process of determining whether a received packet is lost is checked whether or not a frame stored therein is lost (step 320).

This operation is performed in the lost frame detector 110, and checks whether or not a loss is made through a header checksum stored in a header area of the received packet. Whenever a header is created or modified, a bit in the header is checked through cyclic redundancy checking (CRC). When a packet is sent, the same CRC process is repeated in the header. If the result is the same, all the bits in the header are sent correctly, and if different, some of the bits in the header are not sent, which means that some packets were lost during transmission. After all, if it is determined that the packet is lost through the checksum, it is determined that the frame stored in the data area is lost.

If the received current frame is lost, it is determined whether a past frame that has been input in time prior to the received current frame is lost (step 322). The loss of the past frame is not determined through a header checksum as in step 320, but is determined by reading a buffer index (BFI) in which information on whether the frame has been received in the past is stored. The BFI stores information such as loss or lossless for each frame. If it is determined that the previous frame is also lost through the step 322, a first loss restoration method executed in the first restoration module is performed (step 324). In this case, the lost restoration frame is not a current loss frame but a lost past frame. to be. On the other hand, if the past frame is not lost, information about the lost frame is set in the BFI for the currently received frame (step 326). The first loss recovery method (step 324) will be described later in detail.

After the step 326, a new packet is received (step 310), and whether or not the frame stored in the received new packet is lost (step 320). This is to determine whether future frames can be used by judging whether they are lost continuously.

For example, when a lost A _n frame is input when the bitstream frame as shown in FIG. 2 is input, since A _n is lost, the process proceeds to a step 322 of determining whether a past frame is lost, and the previous frame A _{n− 1} is not lost, and the process proceeds to step 326 of storing information in the BFI that the A _n frame is lost. After receiving the packet 310 again, and determines whether the next frame A _{n + 1} is lost or not, since A _{n + 1 is} also a lost frame, the process proceeds to the step 322 of determining whether the previous frame is lost. Since the previous frame A _n is determined to be lost according to the value stored in the BFI, the process proceeds to step 324 of applying the first loss recovery method. In this case, the frame to which the first loss restoration method is applied is an A _n frame, which is a frame to which the loss restoration method has not yet been applied, and according to the above determination result, the A _{n + 1} frame is also lost and cannot be used. The first loss recovery method using only past frames is applied.

On the other hand, if it is determined in step 320 whether the current frame is lost or not, the process proceeds to step 330 in which the previous frame is lost or not. This step 330 also reads the BFI value as in the previous decision step 324 and determines whether a past frame is lost. If there is no loss of the past frame as a result of the determination, the normal decoding step 334 is performed, the lossless information is stored in the BFI for the current frame (step 336), and the data is temporarily stored in the past frame buffer (step 338). ), To receive another packet. However, if there is a loss of the past frame, the second loss restoration method executed in the second restoration module is applied to conceal the loss of the past frame (step 332).

For example, referring again to Figure 2 A _{n +} when the _second frame is input do not have loss of the current frame gamyeo passed to step determines the loss if the past frame 330 (step 320), past frame of A _{n + 1} Since this has been lost, the method proceeds to the second loss recovery method. In the second loss restoration method, A _{n + 1} is restored using the future frame A _{n + 2 that} is not lost and the past frames A _n-2 , A _{n-1 that} are not lost.

Now, the first loss restoration method and the second loss restoration method occurring in the first and second restoration modules will be described with reference to FIGS. 4 to 8.

4 is a flowchart illustrating the first loss restoration method occurring in the first restoration module.

Referring to FIG. 4, a step 410 of retrieving a pitch from frame data stored in the frame buffer 130 before loss, an interpolation 420 of lost frames, an overlap-and-add (OLA) and a scaling step ( Performing step 430 will be described for each step.

The pitch search step 410 determines the pitch as shown in FIG. 5 by finding peak points of normalized cross-correlation from a past frame to obtain a pitch from a frame encoded by the G.711 standard.

The pitch refers to a periodic vibration generated when a voice signal, especially a voiced sound, is expressed and is represented by a time interval at a point indicating a peak value in a periodically repeated voice signal graph. Although it is inherent in characteristics, periodic vibration characteristics such as pitch may appear in an audio signal including a voice signal as in the present invention. That is, the audio signal contains more noise components than the voice signal, but since the sound basically has the repetitive periodicity of the same signal, it is meaningful to apply the pitch search. In addition, the interval for applying the pitch for use in error concealment is within 3 or 4 frames, and if it is more than that, it will be treated as muted. The goal is to conceal errors in loss situations, not to create them. In the absence of any bitstream to decode, the pitch information of the previous frame is very useful, and if you create continuity with the previous signal, It can be prevented from occurring.

Accordingly, in the pitch retrieval step 410 of the first loss reconstruction method 324, a common feature of pre-loss frames is used to recover a frame to replace a lost frame from pre-loss frames stored in the pre-loss frame buffer 130. Search for pitch.

5 is a diagram for describing the pitch search step 410.

Referring to FIG. 5, the pitch search step 410 is described in detail. First, a section in which a value is maximized is determined by a formula of cross correlation between frames before loss, and the section is determined as a pitch.

는 i 번째 샘플의 교차 상관도 값을 의미하며, B 프레임을 i 샘플만큼 왼쪽으로 이동했을 때, A 프레임과 얼마나 많이 닮았는지를 나타내는 지수이다. 이때, B 프레임의 모든 샘플을 검색대상으로 하는 것은 아니며, 일정 영역을 검색 영역으로 지정한다. In Equation 1, A and B are pre-loss frames, respectively, and A is the frame closest in time to the lost frame among the pre-loss frames. Aj means the j th sample of the 1024 samples of the A frame, Bij means the i + j th sample of the 1024 samples of the B frame. In FIG. 5, A and B frames, which are pre-loss frames, are illustrated. The A frame is copied and the equation of the cross correlation with the B frame is applied. remind

Denotes the cross-correlation value of the i th sample, and is an index indicating how much resembles the A frame when the B frame is moved left by the i sample. At this time, not all samples of the B frame are to be searched, and a predetermined area is designated as a search area.

In the present invention, as described in Equation 1, the search is performed based on 160 samples. This is determined in consideration of the condition that the search area should include at least one pitch or most of the pitch, and the problem that if the search area is too long, the calculation amount increases and a period corresponding to n times the pitch may be found. Therefore, 160 samples set an appropriate threshold value based on the value of the pitch appearing in the voice signal. Therefore, a value of the maximum cross correlation is searched for as many as 160 samples corresponding to the search area from the first left sample of the B sample. In FIG. 5, the search region is set in the i-th sample of B and searched, and the maximum correlation point appears at the j-th point of the search region.

On the other hand, as the maximum correlation point appears at the j th point, the most recent data of the previous frames can be regarded as the most similar data before the (b) region, and thus, the (b) region is the pitch including the pitch. . By repeatedly using the pitch retrieved from the frames A and B before the loss, the lost frame is replaced.

At this time, prior to repeating the pitch, OLA is performed in order to ensure continuity in the connection between the pre-loss frame and the pitch to be copied after the pre-loss frame.

Referring back to FIG. 5, when the searched pitch is repeatedly applied, that is, when the (B) region is copied to the rear of the (B) region as it is, the continuity may be degraded and annoying sound may be heard. The previous part of the area (a) and the rear part of the area (b) that overlap with the rear part of the area (b) are overlapped and added to reproduce the frame before the loss.

As shown in FIG. 5, a portion of the (a) quarter of the back of the region and one quarter of the back of the (b) region are selected to perform OLA through a triangular window. The OLA is a method of providing continuity between heterogeneous signals by overlapping data of a part of the front part and the back part of the continuous part. In FIG. 5, the inclination of the hypotenuse of the triangular window is applied to the data of 1/4 pitches for the data at the back of the (a) area and the data at the back of the (b) area expected to be adjacent according to the repeated characteristics of the pitch. OLA is applied while applying the weight, so that the frame before loss is reproduced.

 On the other hand, the pitch search step 410, because a large amount of calculation is required, the complexity of the system is increased. In the audio encoder of the present invention, since the sampling frequency is higher than that of the speech encoder and the number of samples per frame (the typical encoder is 8Khz, one frame is 80 samples, the audio encoder of the present invention is 44.1Khz, One frame is 1024 samples), and as the computational amount increases in this process, the system complexity increases.

Therefore, the search area can be reduced to 8: 1 to reduce the amount of computation. This can reduce computation, but can detect pitches that are two to eight times larger than the actual pitch value, resulting in lower accuracy. Therefore, a method of retrieving the exact pitch period may be obtained by dividing the found pitch value by eight to obtain eight candidate pitches that may become pitches, and then obtaining cross correlations for the eight samples around each candidate pitch. Using this method, more accurate pitch values can be obtained while reducing the complexity of calculation.

After the pitch searching step 410, the loss frame interpolation step 420 is performed.

The generation of the synthesized signal of the lost frame applies the method of G.711 Appendix I based on the pitch retrieved through the search step 410. The first lost frame uses the retrieved pitch once. After that, in the second consecutive lost frames, the pitch used for the synthesized signal is doubled and repeated. In the third consecutive lost frames, the pitch period is added one more time and repeated three times. Subsequent losses are made without further modification. This method serves to attenuate the artificial sound due to the randomly generated pitch when continuous loss occurs.

The lossy frame interpolation step 420 is followed by the OLA and scaling step 430.

The OLA is largely composed of two parts. As described with reference to FIG. 5, in the case of reproducing the pre-loss frame by performing OLA in preparation for connection to the pitch for the pre-loss frame, OLA can be done in the connection section.

FIG. 6 is a diagram illustrating a method for performing OLA at a connection portion of a frame and a pitch before loss.

Referring to FIG. 6, an OLA is performed by applying a triangular window to the data of the rear part of the frame before the loss and the data of the front part of the pitch, respectively, during the overlap length of the predetermined length. The last 128 samples (approximately 2.75 ms at 44.1Khz) of the non-output audio signal of the frame buffer before loss and the 128 samples before the reconstructed signal are weighted together using a triangular window.

On the other hand, the pitch that replaces the lost frame is scaled down as described in ANSI T1.521a-1000 (Annex B). The scale is initially set to 1.0, but if a loss occurs, it is reduced by 0.054 per 10ms from the beginning of the pitch to the quarter frame. From a loss of more than a quarter frame, this decreases by 0.222 per 10 ms. This process continues until the last signal or scale of the output frame is zero.

Now, the second loss restoration method 332 performed in the second restoration module 154 will be described.

7 is a flowchart illustrating the second loss restoration method occurring in the second restoration module.

Referring to FIG. 7, first, the pitch is retrieved from pre-loss frames stored in the pre-loss frame buffer 130 (step 710), and the pitch is retrieved from the post-loss frames stored in the frame buffer 140 after loss (step). 720, the lost frames are interpolated using the searched two pitches (step 730), and the OLA and scaling steps 740 are performed on the generated lost frames.

First, referring to the pitch search steps 710 and 720, the step 710 of searching for the pitch from the frames before loss uses the same method as the pitch search step 410 performed in the first reconstruction module.

The step 720 of retrieving the pitch from the frames after loss is also not significantly different from the principle of the pitch retrieval step 410. In order to obtain the pitch from the frames after the loss, the first frame of two consecutive frames after the loss is input, the B frame of Equation 1, and the second frame as the A frame to search for the pitch.

In order to provide continuity for the connection between the retrieved pitch and the lost frame, the frame after the loss is reproduced by applying the OLA by one quarter pitch of the signal after one pitch period of the lost frame and the start signal of the lost frame.

The pitch obtained from the pre-lost buffer and the pitch from the post-lost buffer are passed to lost frame interpolation step 730 to recover lost packets, respectively.

The lost frame interpolation step 730 applies the method of G.711 Appendix I as in the lost frame interpolation step 410 in the first reconstruction module. The pitch retrieved from the frame before the loss is repeated and the pitch retrieved from the frame after the loss is repeated to replace the lost frame.

After the interpolation step 730, OLA and scaling are applied between the two pitches replacing the lost frame (step 740).

FIG. 8 is a diagram illustrating the step of applying an OLA in the connection portion of the pitch obtained from the pre-loss buffer and the pitch obtained from the post-loss buffer.

Referring to FIG. 8, an OLA is applied to a packet generated using a post-loss signal and a packet generated using a pre-loss signal. In this case, the overlap length to which the OLA is applied depends on the length of the lost frame as shown in Equation 2.

Overlap Length = (Pitch period retrieved in frames after loss) / 4 + erase_cnt × 128 (2.75 ms)

{overlap length, <1024 samples (1 frame)

{1024 samples,> = 1024 samples (1 frame)

The erase_cnt is a value that counts the number of consecutively lost frames. As the number of consecutive losses increases, the correlation length between the frames before the loss and the frames after the loss becomes smaller, so that the overlap length is applied.

As in the OLA process described with reference to FIG. 6, an OLA is performed by applying a triangular window to each of the pitch obtained from the pre-loss frame buffer and the pitch obtained from the post-loss buffer.

Meanwhile, as in the scaling step 430 of the first loss recovery method, the pitch replacing the lost frame is scaled down. The scale is initially set to 1.0, but if a loss occurs, it is reduced by 0.054 per 10ms from the start of the frame to the quarter frame. And from a loss of more than 1/4 frame, it decreases by 0.222 per 10ms. This process continues until the last signal or scale of the output frame is zero.

The present invention restores a lost frame with a frame that was normally transmitted before the loss if the frame after the loss is not available, and if the packet after the loss is available, the frame and the normal transmission before the loss By reconstructing the frame after the loss, the sound quality degradation can be reduced by concealing the lost frame when the loss of the audio frame occurs. In addition, the method does not require a lot of computational process, and since the frames are used before and after lost packets, the accuracy of restoration can be improved. In particular, it can be said that it is suitable for a broadcasting system such as DMB which does not have a large number of continuous frame loss.

Claims (9)

Detecting whether an audio frame stored in the received audio packet is lost; Detecting whether the audio frame is lost when the audio frame is lost; A loss retrieval step of retrieving a pitch from pre-loss frames and replacing the lost frame by repeating the retrieved pitch when the audio frame is continuously lost Audio packet loss recovery method comprising the. The method of claim 1, wherein the loss recovery step includes repeating the retrieved pitch one or more times according to the number of lost frames; Performing overlap-and-add (OLA) on the connection between the pre-loss frame and the pitch; Scaling down the connected pitch The audio packet loss recovery method further comprises. The method of claim 1, wherein the audio signal stored in the received audio packet is sampled according to a sampling frequency of 44.1 Khz, and the audio frame includes 1024 samples. Detecting whether an audio frame stored in the received audio packet is lost; Detecting whether the audio frame is lost when the audio frame is lost; A loss retrieval step of retrieving a pitch frame from pre-loss frames and post-loss frames if a lossless frame is input immediately after the loss of the audio frame, and repeating the retrieved pitch frames to replace the lost frame. Audio packet loss recovery method comprising the. 5. The method of claim 4, wherein the loss recovery step includes repeating the retrieved pitch frames one or more times according to the number of lost frames; Performing overlap-and-add (OLA) on the pitch frame retrieved from the frames before the loss and the pitch frame retrieved from the frames after the loss; Scaling down the connected pitch frame The audio packet loss recovery method further comprises. A lost frame detector for detecting whether an audio frame stored in the received audio packet is lost; A decoder for decoding the received audio frame; A pre-loss frame buffer for temporarily storing frames before the lost frame; A post-loss frame buffer for temporarily storing frames after the lost frame; A first reconstruction module for replacing a pitch retrieved from the pre-loss frame buffer with a lost frame; A second reconstruction module that combines the pitch retrieved from the pre-loss frame buffer with the pitch retrieved from the post-loss frame buffer and replaces the lost frame with a lost frame Audio packet loss recovery device comprising a. 7. The apparatus of claim 6, wherein the decoder complies with the MPEG-4 Bit Sliced Arithmetic Coding (BSAC) standard. 7. The apparatus of claim 6, wherein the first reconstruction module executes overlap-and-add (OLA) in the connection portion of the pre-loss frame and the pitch. 7. The apparatus of claim 6, wherein the second reconstruction module implements overlap-and-add (OLA) to combine the pitches.

Images