JP2002221994A - Method and apparatus for assembling packet of code string of voice signal, method and apparatus for disassembling packet, program for executing these methods, and recording medium for recording program thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress the deterioration of voice quality caused by packet dissipa tion by adding a small amount of auxiliary information. SOLUTION: In assembling of voice packets, voice signals are coded for every frame and code string are created. The code strings (high quality main bit stream) (1), etc., of the current frame is combined with the code strings (subbit stream compressed at a high compression rate) (2), etc., of a previous frame, respectively, and then the voice packets (1), etc., of the current frame time are created and stored. In the disassembling of the voice packets, the code strings (main bit stream) of the current frame are decoded among the stored code strings when voice packets do not dissipate. When voice packets (3) dissipate, the code strings (3) (subbit stream) of the previous frame of the voice packets (2) are decoded to reproduce the voice signals.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for assembling a packet of a code string of an audio signal used in a technique for suppressing and compensating for quality deterioration of an audio signal lost due to packet loss which may occur when transmitting an audio signal in packets. The present invention relates to a packet disassembly method and apparatus, a program for executing these methods, and a storage medium for storing the program.

[0002]

2. Description of the Related Art As represented by mobile communication and VoIP (Voice over IP), voice and data can be integratedly handled by packet communication. In many cases, a plurality of voice frames are packed and transmitted in one packet in order to reduce the overhead of the header of the normal packet. In voice communication, in general, it is essential that the order in which a signal sequence constituting a voice signal is generated is maintained even when the signal is reproduced after transmission, but in packet communication, transmission of packets transmitted at regular intervals is required. Since the delay varies, the arrival time may fluctuate. A fluctuation absorbing buffer is used to reproduce the audio signal by the code stored in the packet in the order in which the fluctuation is absorbed and transmitted.

One of the problems in packet voice communication is packet loss. By increasing the bandwidth and speed of the communication path, the degradation and delay due to coding are eliminated. On the other hand, packet loss is a problem that cannot be avoided even if the communication capacity increases. The causes of packet loss include the following. First, if the number of packets is large,
A packet may be completely lost due to collision between packets. When the code bit error reaches about 50% in the course of transmission, it is assumed that all the packet information has been lost, and it may be determined that the packet has been lost. Furthermore, if the arrival delay of a packet is larger than compensated by the fluctuation absorbing buffer, it may be determined that the packet has been lost and that the packet has been lost. These causes the voice quality to deteriorate.

[0004] In order not to make the hearing uncomfortable due to the deterioration of quality, the part of the lost packet needs to be compensated by some other signal. Depending on the encoding method, encoding is performed using information before and after the buffer. Therefore, once a packet is lost, quality may deteriorate for a while after restoration. Suppressing the quality deterioration from the viewpoint of hearing is also included in the packet loss compensation. With reference to FIG. 9, a description will be given of a conventional technique for suppressing and compensating for quality degradation of a voice signal lost due to packet loss.

[0005] The missing voice interpolation device comprises a demultiplexing circuit 102,
Residual signal power decoder 119, inverse quantizer 120, long-term prediction coefficient decoding / selector 118, short-term prediction coefficient decoder 107, long-time synthesis / interpolation filter 117, short-time synthesis filter 110, and input / output terminals 101, 115, 116 When the loss of the voice coded information is detected, that is, when the loss detection signal is input, the connection cutoff switch 121 is opened and the long-term prediction coefficient decoding / selector 118 Temporal prediction coefficient (1. Numerical value is set in advance, always constant value, or 2. Long-term prediction coefficient for interpolation according to long-term prediction coefficient of previous frame obtained from long-term prediction coefficient decoder) It is output to the long-term predictor 112. In the short-time predictor 114, the short-time prediction coefficient in the previous frame is set as it is. The self-driving without any input to the long-time synthesis / interpolation filter 117 allows the output signal to be
To enter. The short-time synthesis filter 110 performs normal decoding processing to interpolate the reproduced digital audio signal. This conventional technique is a method of generating a pseudo signal by analyzing a pitch period from a past signal, extracting an appropriate waveform, and repeating the same in a decoder. . The most likely cause of the deterioration in the pitch cycle repetition compensation is the discontinuity of the waveform. Discontinuities in the waveform are likely to occur in a portion where the compensation signal during packet loss and the signal after recovery from packet loss are joined.
In order to make this discontinuity inconspicuous, the pitch cycle is adjusted so that it is continuous with that after returning from disappearance, or
A method of gradually changing the combined signal and the restored signal by OLA (Overlap add) and gradually attenuating the power of the combined signal have been proposed.

Used for low bit rate speech coding
In CELP (Code Excited Linear Prediction) packet erasure compensation, the audio signal in a packet is classified into periodic and aperiodic in advance, and if the pitch frequency of the lost packet is periodic, A method of using an excitation signal of an adaptive codebook and randomly using white noise if it is aperiodic is often used. Still other methods include, as characteristic processes, a method of repeating a synthesis filter coefficient, attenuating adaptive / fixed codebook gain, and attenuating gain prediction.

[0007] These methods are effective methods for suppressing an unpleasant signal for hearing. However, in many cases, it is difficult to reproduce a sound that is close to the original sound because the reproduction is a pseudo synthetic signal. Between packets, the pitch or power may fluctuate rapidly, or the sound quality may be significantly degraded due to waveform discontinuity or unreasonable adjustment due to pitch interval mismatch. Further, in the case of the compression codec, there is a problem that the rising portion is deteriorated after recovery from the disappearance.

[0008]

SUMMARY OF THE INVENTION The present invention solves the drawbacks of the conventional packet loss compensation technique, assuming that the capacity of the communication path is sufficiently large and some auxiliary information can be added. The task is to improve the deterioration. In the prior art, when the pitch period, the power, and the like change in the section where the packet has been lost, the deterioration becomes significant. The present invention provides an audio signal while a packet is lost, and an encoding and decoding method capable of suppressing deterioration of an audio signal after restoration from loss, even if the audio data length included in the packet is large, and It is an object to provide means for realizing these methods.

[0009]

In order to solve the above-mentioned problems, the present invention provides a method of assembling a voice packet, in which a voice signal is encoded for each frame to generate a code sequence, and the code of the current frame and the previous frame is generated. The columns are combined to create and store the packet at the current frame time, and in the disassembly of the voice packet, if no packet has been lost, of the code string stored in the packet at the current frame time, The code string is decoded, and if the packet is lost, the code string of the current frame of the packet at the previous frame time is decoded to reproduce the audio signal.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS As shown in FIG.
In (r Packet), a voice packet is received by the network module 1, decomposed for each packet and output to the fluctuation absorbing buffer 3, and packet loss is determined and a packet loss flag is output to the packet loss compensator 2. The decomposed packets are stored in the fluctuation absorbing buffer 3, and the packets are reproduced after waiting for a while. If the packet loss is not determined, the main bit stream of the current frame of the packet stored in the fluctuation absorbing buffer 3 is output to the audio decoder 4. When packet loss is determined, efficient packet loss compensation can be performed using the main and sub bit streams of the packets before and after reaching the fluctuation absorbing buffer 3. For example, when a packet to be reproduced does not arrive, the audio decoder 4
Creates an analysis coefficient and outputs it to the packet loss compensator 2,
The packet erasure compensator 2 outputs the erasure compensation data to the audio decoder 4 using the analysis coefficient and the sub-bitstream of the past packet that has reached the fluctuation absorbing buffer 3, and can perform efficient packet erasure compensation. .
The audio decoder 4 performs processing to suppress deterioration as much as possible by performing parameter interpolation and volume control, and can output audio.

In the present invention, one main codec which is normally used is provided with resistance to packet loss by combining a plurality of sub-codecs as a compensating means when a packet is lost. The main codec uses a high-quality coding method with a relatively low compression rate, and the sub-codec needs to select a coding method with higher compression and higher quality than the main codec. By doing so, an increase in the data amount can be suppressed. (Encoder) The encoder of the present invention will be described with reference to FIGS.

The input audio data is divided into frames (data units), and a signal to be normally reproduced on the decoder side is encoded by a main codec (encoder) 11 by a first encoding method to form a main bit stream. . For example, encoding is performed according to the G.711 encoding standard, which is a pulse code modulation (PCM) system that transmits an 8 kHz sampling audio band signal at 64 kb / s in the audio encoding system recommended by the ITU-T. In addition, the sub-codec (encoder) 14 encodes the audio data one packet ahead by the second encoding method to create a sub-bit stream. For example, encoding is performed according to the G.728 encoding standard, which is an LD-CELP scheme for 16 kb / s telephone band audio encoding recommended by ITU-T. When two or more sub-bitstreams are included, it is preferable to include a bitstream obtained by encoding a signal that is pre-read more than that.

The main bit stream and the sub-bit stream one packet ahead are combined by a combining unit 15, and a process of adding a chronological sequence number, a bit error detection code, and the like by a known serial number assigning circuit or the like is performed by a packetizing unit. The packet is output at step 16 and packetized. In the above-described manner, each packet has a main bit stream in which the signal stored in the main buffer is encoded by the main codec, and a sub-bit in which the signal ahead of the main buffer stored in the sub buffer is encoded by the sub codec. Have a stream.

As shown in FIG. 4, by creating a bit stream in this way, when the decoding side determines that a packet has been lost (packet), the sub-bit stream included in the packet immediately before the loss is determined. , The speech (decoded speech signal) of the lost section can be created. This is because the sub-bit stream includes a signal obtained by encoding a pre-read signal using a sub-codec. In addition, in order to reduce the overhead of the packet header, a plurality of voice frames are often packed in one packet and transmitted. In the conventional packet erasure compensation, a pseudo audio signal is generated by repeating a past signal using a synthesized voice, so that the longer the audio data included in the packet is, the more remarkable the deterioration becomes.

In the present invention, since the sub codec includes audio data one packet ahead of the main codec, packet loss compensation with little deterioration can be performed regardless of the length of the audio data included in each packet. . Care must be taken depending on the compression coding used for the sub-codec. It is the analysis coefficient required for encoding and decoding (this depends on the codec,
For example, a codec that inherits and decodes a synthesized signal, a filter coefficient, a prediction coefficient, and the like from a previous packet, and decodes the analysis coefficient of a predictor or a quantizer by a normal encoder and decoder when a packet loss occurs. Will be different. Even in such a case, in order to match the analysis coefficients, it is necessary for the encoder to transmit initial information of the analysis coefficients as encoded information.

In the present invention, a main bit stream coded by a main codec, which is high quality coding, and a sub bit stream coded by a sub codec, which is high compression coding, are present as a set inside a packet. I do. Therefore, if the analysis coefficient of the sub codec is created by the main codec (decoder) 12 and the sub codec (analysis coefficient calculation) 13 from the signal decoded by the main codec, the information need not be transmitted. For example, G.
When the analysis coefficients of the encoder are generated from past synthesized signals as in 728, it is possible to replace the synthesized signal portion with a high-quality signal decoded by the main codec. Similarly, in the decoder, it is necessary to replace the synthesized signal portion with high quality coding. The decoding can be performed correctly by matching the internal states of the encoder and the decoder. Further, the quality of the decoded signal can be improved by replacing the analysis coefficient with a high-quality signal instead of the synthesized signal.

The analysis coefficients refer to, for example, a synthesis filter coefficient obtained by a backward synthesis filter adaptor of a G.728: LD-CELP encoder (not shown) and an auditory weighting filter coefficient obtained by an auditory weighting filter. The synthetic inverted filter _{e n = x n + a 1} x n over _{1 + ··· + a n x 0} a: filter coefficient x: synthesized signal e: to replace higher-quality signal to perceptual weighting filter in a manner similar residual signal Is also possible.

An auditory weighting filter w_n= A₀x_n+ A₁x_nー₁+ ... + a_nx_o − (B₀w_n+ B
₁w_nー₁+ ... + b_nw_oa, b: filter coefficient x: synthesized signal w: auditory weighting
Signal in this way, by replacing it with a higher quality signal.
High-quality decoding. Calculated minutes
The analysis coefficients are transferred to the sub codec (decoder) 14 and
That is, G.728: Optimal code of LD-CELP encoder (not shown)
Codebook (sign) as output from book data selector
Code book (waveform) stored in the
The code corresponding to the best one is selected from the gain levels.
And a sub-bitstream is output. (Decoder) The decoder will be described with reference to FIGS.

On the decoder side, first, the received signal is depacketized by the depacketizing section 21 and the main codec /
In the sub-codec distribution unit, the main bit stream (G.711
It is distributed to a code sequence according to the audio coding standard) and a sub-bitstream (code sequence according to the G.728 audio coding standard of the previous frame). The main bit stream is decoded by the main codec (decoder) 23 into an audio signal by the first decoding method.

Then, the decoded signal is used to calculate an analysis coefficient by a sub codec (analysis coefficient calculation) 24 to create an internal state of the sub codec (decoder) 25. Alternatively, as described above, the internal state of the sub codec is created directly from the main codec. Finally, with the internal state being inherited, the sub-codestream (decoder) 25 decodes the sub-bit stream by the second decoding method, and outputs an audio signal.

Specifically, an LD-CELP decoder (not shown)
Input the code for the shape code vector (waveform) and the code for the gain level as a coded sequence according to the G.728 speech coding standard, select the shape code vector and the gain vector from the codebook (excitation VW codebook), and The decoded speech is reproduced by transferring the analysis coefficients calculated as the filter coefficients in the synthesis filter. The detection of the presence / absence of the voice packet loss is performed before the depacketizing unit 21 shown in FIG. 5, and is performed by detecting a disorder of the sequence number or a bit error by a general-purpose packet loss detection circuit.

If it is not determined that there is a packet loss signal, the changeover switch 27 is switched to the main codec (decoder) 23 to output an audio signal. When it is determined that the packet loss signal is present, the changeover switch 2
7 is switched to the sub codec (decoder) 25 side.
The main codec is ADPCM (Adaptive Pulse Code Modula)
If the main codec requires past information, such as (tion), that is, a codec that takes over the internal state, and the past packet is lost, the sub codec used for erasure compensation is decoded. Deterioration occurs in the connection between the decoded signal and the audio signal decoded from the main codec. In such a case, by creating necessary information from the signal reproduced by the sub codec, it is possible to suppress the deterioration of the reproduction of the main codec after the compensation.

When there is only one sub-codec and packets are continuously lost, if there are not as many sub-codecs as the number of lost packets, erasure compensation by the sub-codec cannot be performed, and if the sound is degraded. Conceivable. In such a case, the compensation shown in FIG. 6 is performed by using the waveform repetition compensator 26 according to the conventional method shown in FIG. Only when the sub-codec cannot be used, a waveform is generated using a composite signal such as the past pitch frequency repetition erasure compensation. As a countermeasure for the case of burst loss (two or more packet losses), an example in which compensation is performed by a conventional method as shown in FIG. 5 is shown. It is sufficient to store a code sequence based on a signal, and decode a speech signal using a code sequence in a preceding frame from a code sequence extending over two or more frames in packet decomposition.

If the main codec and the sub codec have no great difference in the degree of distortion of the quantization noise, the signal-to-quantization noise ratio (S
NR: Signal-to-quantization Noise Ratio). That is why quantization noise is often uncorrelated for different codecs, and the sum of the powers of the correlated and uncorrelated noise is considered to be greater in the acoustic part when added together. . As the number of sub-codecs increases, the more pre-read information is provided for the main codec. As a result, it is possible to provide resistance even when packets are continuously lost.

However, as shown in FIG. 7, as the number of sub-codecs is increased, pre-read information is required, and as a result, the delay is increased. Also. The amount of information increases by the number of sub-codecs. The cause of the delay in VoIP is a large delay caused by a fluctuation absorbing buffer that absorbs fluctuation such as packet arrival delay other than the above. In addition, a large delay occurs on a PC (Personal Computer) due to a buffer of a network card, a sound card, or the like, but this is solved by introducing dedicated hardware or improving the performance of the PC. For real-time conversations, it is desirable that the total delay time in one direction be within 200 milliseconds. Therefore, it is necessary to adjust the total number of sub-codecs, fluctuation absorption, and other delays to meet the standard.

In mobile communication and VoIP, the communication speed is not always constant, and it is considered that the amount of information that can be used for voice communication changes depending on the amount of information used for applications. The present invention is characterized in that a combination suitable for a network is enabled by flexibly changing the quality of a sub-codec and the number of sub-codecs according to a communication speed and a calculation speed of a computer. FIG. 8 shows a schematic diagram of a waveform when this method is used. Referring to this figure, it can be seen that the present method is closer to the original sound than the conventional method.

The packet assembling apparatus and the packet disassembling apparatus according to the present invention may include a computer having a CPU, a memory, and the like.
User terminal and CD used by the user who will be the subject of access
-It can be composed of a machine-readable recording medium such as a ROM, a magnetic disk device, and a semiconductor memory. A control program for causing a computer to execute the above-described operation is stored in a recording medium, and the control program is read by a computer, and the operation of the computer is controlled to realize each element in the above-described embodiment on the computer. it can.

[0028]

According to the present invention, as compared with the conventional method, deterioration in quality due to packet loss can be suppressed as much as possible, and discontinuous portions of the waveform can be eliminated, and the lost portion faithful to the original sound can be compensated. . Further, since the code sequence of the current frame and the code sequence of the previous frame are combined as the packet at the current frame time, even if the previous packet is lost, it can be easily compensated.

[Brief description of the drawings]

FIG. 1 is a diagram showing a basic configuration of VoP.

FIG. 2 is an explanatory diagram of processing of an encoder.

FIG. 3 is a schematic configuration diagram of an encoder.

FIG. 4 is an explanatory diagram of processing when a packet is lost.

FIG. 5 is a schematic configuration diagram of a decoder.

FIG. 6 is an explanatory diagram of a process when a burst is lost.

FIG. 7 is a diagram showing the structure of one packet when a plurality of sub-codecs are provided.

FIG. 8 is a comparison diagram of waveforms of the prior art and the method of the present invention for original sound.

FIG. 9 is a configuration diagram of a conventional missing voice interpolation device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Network module 2 Packet loss compensation part 3 Fluctuation absorption buffer 4 Audio decoder 11 Main codec (encoder) 12, 23 Main codec (decoder) 13, 24 Sub codec (calculation of analysis coefficient) 14 Sub codec (encoder) 15 Coupling part 16 packet Conversion unit 21 depacketization unit 22 main codec / sub-codec distribution unit 25 sub-codec (decoder) 26 waveform repetition compensation unit 27 switch

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5D045 DA20 5J064 AA01 BB04 BC01 BC02 BD02 BD03 BD04 5K067 AA23 BB04 CC08 EE02 GG11 HH21 HH23

Claims (19)

[Claims] 1. A packet assembling method for assembling a packet of a code string of an audio signal, comprising the steps of: encoding an audio signal for each frame to generate a code string; Combining and storing code strings of N (N: any integer) frames. 2. A packet assembling method for assembling a packet of a code string of an audio signal, wherein the audio signal is encoded for each frame by a high-quality first encoding method. Generating a second code sequence by encoding a speech signal of a preceding N (N: any integer) frame by a second encoding method with high compression for each frame; Combining and storing the first code string of the current frame and the second code string of the previous frame as a packet of the current frame time. 3. The packet assembling method according to claim 2, wherein the step of generating the second code sequence includes the steps of: decoding from the first code sequence to generate a decoded signal; Calculating an analysis coefficient, wherein the second encoding method includes the step of generating the second code sequence from the audio signal using the analysis coefficient. Method. 4. A packet assembling apparatus for assembling a packet of a code sequence of an audio signal, comprising: an encoder for encoding the audio signal for each frame to generate a code sequence; A combining unit for combining and storing code strings of N (N: any integer) frames. 5. A packet assembling apparatus for assembling a packet of a code string of an audio signal, wherein the audio signal is encoded by a high-quality first encoding method for the audio signal for each frame, and the first code string is encoded. A first encoder that generates a second code sequence by coding an audio signal for each frame with a code sequence of a preceding N (N: an arbitrary integer) frame using a high-compression second coding method; A second encoder, combining the code sequence of the first frame of the current frame generated by the first encoder as the packet of the current frame time and the second code sequence of the previous frame generated by the second encoder A packet assembling device, comprising: a coupling unit for storing. 6. The packet assembling apparatus according to claim 5, further comprising: means for decoding from the first code sequence to generate a decoded signal; and means for calculating an analysis coefficient from the decoded signal. A second encoding method for a second encoder that generates a sequence, comprising: means for generating the second code sequence from the audio signal using an analysis coefficient. 7. A packet at the current frame time obtained by combining the code sequence of the current frame and the code sequence of the previous N (N: an arbitrary integer) frame is input, and the speech is generated from the code sequence stored in the packet for each frame. In a packet disassembly method for decoding a signal, a process of determining whether or not a packet has been lost for each frame; and, if it is determined that the packet at the current frame time has not been lost, the current frame stored in the packet at the current frame time. Decoding the audio signal from the code string of the above, and decoding the audio signal from the code string of the current frame included in the packet at the past frame time when it is determined that the packet at the current frame time is lost. A packet disassembly method comprising: 8. An audio signal is encoded for each frame by a high-quality first encoding method using a code sequence of a current frame to generate a first code sequence. : Arbitrary integer) The code sequence of the frame is encoded by the second encoding method of high compression to generate a second code sequence, and the first code sequence of the current frame is generated.
A packet disassembly method for inputting a packet at the current frame time obtained by combining the code sequence of the frame of the previous frame and the second code sequence of the previous frame and decoding the audio signal from the code sequence stored in the packet for each frame, Determining whether or not the packet at the current frame time has not been lost; and determining from the first code sequence of the current frame among the code sequences stored in the packet at the current frame time when it is determined that the packet at the current frame time has not been lost. A first step of decoding an audio signal by a first decoding method corresponding to the first encoding method, and a packet of a past frame time when it is determined that the packet at the current frame time is lost. A second step of decoding an audio signal from a second code string of the current frame among the code strings stored in the second frame by a second coding method and a corresponding second decoding method. Packet decomposition method according to symptoms. 9. A packet decomposing method according to claim 8, wherein an analysis coefficient is calculated from an audio signal decoded from a first code sequence of a past frame among code sequences stored in a packet at a past frame time. And in the second step, the second encoding method uses a second code of a current frame in a code string stored in a packet at a past frame time using the analysis coefficient. A packet disassembly method comprising decoding an audio signal from a sequence. 10. The packet disassembly method according to claim 7, wherein a plurality of packets at the current frame time are determined to be continuously lost, and the speech of the current frame is determined from the second code string. Generating a pseudo audio signal from frames before and after the frame that cannot be decoded if the signal cannot be decoded, and performing erasure compensation. 11. The packet disassembly method according to claim 7, wherein when it is determined that the packet at the current frame time has not been lost, the packet is included in the current frame signal and the past packet. A packet disassembly method comprising a step of generating a voice signal by adding a current frame signal. 12. An audio signal is encoded for each frame to generate a code sequence, and a packet at the current frame time obtained by combining the code sequence of the current frame and the code sequence of the preceding N (N: any integer) is input. A packet decomposer that decodes an audio signal and determines whether or not a packet has been lost for each frame; and if the packet loss determination means determines that the packet at the current frame time has not been lost, The audio signal is decoded from the code string of the current frame stored in the packet of the current frame time, and if it is determined that the packet of the current frame time has been lost, the code string of the current frame included in the packet of the previous frame time And a decoding means for decoding an audio signal from the packet signal. 13. An audio signal is encoded for each frame by a high-quality first encoding method using a code sequence of a current frame to generate a first code sequence. (Arbitrary integer) The code sequence of the frame is encoded by the second encoding method of high compression to generate a second code sequence, and the first code sequence of the current frame and the second code sequence of the previous frame are generated. In a packet decomposer for inputting a packet of the combined current frame time and decoding an audio signal from a code string stored in a packet for each frame, a packet loss determining means for determining whether or not a packet is lost for each frame; If the determination means determines that the packet at the current frame time has not been lost, it corresponds to the first encoding method from the first code sequence of the current frame among the code sequences stored in the packet at the current frame time. Do When the audio signal is decoded by the decoding method of No. 1 and it is determined that the packet at the current frame time has been lost, the second code of the current frame in the code string stored in the packet at the previous frame time is used. Decoding means for decoding an audio signal from a sequence by a second decoding method corresponding to the second encoding method. 14. A packet decomposing apparatus according to claim 13, wherein said decoding means comprises a decoding means for converting a speech signal decoded from a first code string of a past frame among code strings stored in a packet at a past frame time. Means for calculating an analysis coefficient, wherein the second encoding method uses the analysis coefficient to convert an audio signal from a second code string of a current frame among code strings stored in a packet at a previous frame time. A packet decomposing device for decoding. 15. The packet disassembly device according to claim 12, wherein the decoding means determines whether the packet at the current frame time has not been lost and the current frame signal and the past frame signal. A packet decomposer for generating an audio signal by adding a current frame signal included in a packet. 16. A computer for encoding a speech signal of a current frame by a high-quality first encoding method to generate a first code string for each frame; Encoding the code sequence of the previous N (N: any integer) frame by the second encoding method of high compression to generate a second code sequence; and the first frame of the current frame as a packet at the current frame time. And a program for executing a procedure of combining and storing the code sequence of the first frame and the second code sequence of the previous frame. 17. A computer comprising: a computer for encoding a speech signal of a current frame by a high-quality first encoding method for each frame to generate a first code string; Encoding the code sequence of the previous N (N: integer) frame by a second encoding method of high compression to generate a second code sequence; and a first frame of the current frame as a packet at the current frame time. A computer-readable recording medium on which a program for executing a procedure for combining and storing the code sequence of the previous frame and the second code sequence of the previous frame is stored. 18. A procedure for inputting a packet at a current frame time obtained by combining a code sequence of a current frame and a code sequence of a previous N (N: an arbitrary integer) frame into a computer, and determining whether or not a packet has been lost for each frame. A determining step; a step of decoding an audio signal from a code string of the current frame stored in the packet at the current frame time when it is determined that the packet at the current frame time has not been lost; Is a program for executing a procedure for decoding an audio signal from a code string of a current frame included in a packet at a past frame time when it is determined that is lost. 19. A procedure for inputting, as a packet at the current frame time, a packet at the current frame time obtained by combining the code string of the current frame and the code string of the previous N (N: an arbitrary integer) frame to the computer; And a procedure for decoding the audio signal from the code string of the current frame stored in the packet at the current frame time when it is determined that the packet at the current frame time has not been lost. If it is determined that the packet at the current frame time has been lost, a computer-readable program that records a program for executing a procedure for decoding the audio signal from the code string of the current frame included in the packet at the past frame time recoding media.