JPH11298448A - Error correction code generating method, error correction processing method, voice coder and voice decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To strengthen resistance to a transmission error while minimizing the increase in a transmission bit rate. SOLUTION: A linear prediction analysis means 101 applies linear prediction analysis to an input voice signal to generate a line spectrum pair(LSP) parameter. A code book parameter generating means 103 generates code book parameters ia, iga, is and igs. A 1 forward error correction(FEC) 105 generates a 4-bit parity flag by a (15, 11) humming code method for an information vector in 11-bit. 2 FEC 106a-106e individually generate a parity flag for the LSP parameter and for each code book parameter respectively by a 1-bit parity code method. The LSP parameter, each code book parameter, a parity bit and each parity flag are multiplexed, and a transmission block is generated and sent to a transmission line 301.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an error correction code generation processing method for generating an error correction code added to a transmitted information block for checking a transmission error, and a reception method comprising an information block and an error correction code. The present invention relates to an error correction processing method for inspecting a transmission error of a transmitted transmission block and correcting or correcting the information block based on the inspection result, and further generating an audio parameter from an input audio signal (encoding the input audio signal) Generating an error correction code for the information block including the voice parameter, transmitting a transmission block including the information block and the error correction code to the voice decoding side, and inspecting a transmission error of the received transmission block. , Corrects or corrects the information block based on the inspection result, and performs the inspection / correction / correction of the audio parameter. (To decode the speech signal testing / correction / compensation is encoded after completion) for reproducing audio signal based on the data about the speech decoding apparatus.

[0002]

2. Description of the Related Art A speech encoder and a speech decoder are used for digital radio communication, voice mail, and the like. As a speech encoding / decoding method used for a pair of speech encoding apparatus and speech decoding apparatus, CELP is used.
(Code Excited Linear Prediction) method. The CELP method is 4.8 to 8.0 [kbp
s (kilo bits per second)], which is a speech encoding / decoding technique suitable for wireless communication at a low bit rate (for example, car phone or mobile satellite communication).

[0003] In the CELP method, a linear prediction parameter (a parameter corresponding to a spectral envelope of an input speech signal) specifying a coefficient of a speech synthesis filter and a codebook parameter designating an optimal excitation vector for driving the speech synthesis filter are used for speech. This is a speech encoding / decoding technique for transmitting as a parameter, wherein a codebook storing excitation vectors and a speech synthesis filter are provided on the speech encoding side and the speech decoding side, respectively. In addition to generating linear prediction parameters by predictive analysis, a codebook parameter is generated by searching for an optimal excitation vector from the codebook using a speech synthesis filter, and the speech decoding side generates a speech synthesis filter according to the linear prediction parameter. Set the coefficients and based on the codebook parameters, The excitation vector retrieved from codebook such, synthesized (play) audio signal by driving the speech synthesis filter by this excitation vector.

[0004] A sound source vector in the CELP method includes an adaptive sound source vector representing a periodic component (pitch) of speech, and
It is composed of a noise sound source vector that expresses a changed part of speech that cannot be expressed by the adaptive sound source vector. The codebook includes an adaptive codebook in which the adaptive excitation vector is stored, an adaptive gain codebook in which the gain for the adaptive excitation vector is stored, a noise codebook in which the noise excitation vector is stored, and a gain for the noise excitation vector. It consists of four kinds of codebooks of the noise gain codebook. The codebook parameters are an adaptive codebook parameter indicating an optimal adaptive excitation vector, an adaptive codebook parameter indicating an optimal adaptive excitation vector gain, a noise codebook parameter indicating an optimal noise excitation vector, and an optimal noise source vector. Four types of noise gain codebook parameters indicating the excitation vector gain are generated.

[0005] In addition, in a pair of speech encoding apparatus and speech decoding apparatus, in order to prevent a decrease in reproduction speech quality due to a transmission error of an information block including various speech parameters, the speech encoding apparatus includes an information block. An error correction code for checking the transmission error of the information block is generated (error correction code generation processing), a transmission block including the information block and the error correction code is transmitted, and the transmission error of the received transmission block is detected by the speech decoding device. By inspecting and correcting or correcting the information block based on the inspection result (error correction processing), the information block is protected from transmission errors. It is necessary to use an error correction method with a high error correction capability for a bad burst transmission path such as a mobile radio, but by adding an error correction code to all bits of an information block by such an error correction method, It is not practical because it involves an increase in the transmission bit rate. on the other hand,
The voice parameter bits include bits that significantly degrade the sound quality when a transmission error occurs, and bits that do not significantly degrade. Therefore, by adding an error correction code using a high error correction technique with a high error correction capability only to bits that significantly deteriorate sound quality when a transmission error occurs, and by not adding an error correction code to bits that do not significantly deteriorate sound quality, Important bits in the information block are intensively protected.

The above error correction method includes (15, 1)
1) There is a Hamming code method. The Hamming coding method is a binary linear coding method capable of correcting a single-bit error.
11) The Hamming code method adds 4 parity bits to an 11-bit information vector,
A 15-bit transmission vector is generated.

FIG. 6 is a block diagram showing a configuration of a conventional speech coding apparatus. 6 includes a linear prediction analysis unit 101 and a subframe linear interpolation unit 102.
Codebook parameter generation means 103, adaptive codebook 1
04a, adaptive gain codebook 104b, and random codebook 10
4c, a noise gain codebook 104d, an error correction code generation means (FEC: Forward Error Correction) 108,
And a multiplexer 109.

The linear prediction analysis means 101 performs a linear prediction analysis on the input speech signal to obtain an LSP for each frame.
Generate (Line Spectrum Pair) parameters. The LSP parameter is a linear prediction parameter having good interpolation characteristics and quantization characteristics. When the human vocal tract is considered as an acoustic tube, a line spectrum frequency corresponding to the resonance characteristic of the vocal tract acoustic tube, that is, a boundary condition is paired. Is obtained as a parameter of. The subframe linear interpolation means 102 generates LSP parameters in subframe units (one obtained by dividing a frame into four) by interpolating LSP parameters in frame units, and further converts the LSP parameters into filter coefficients. Is set in the speech synthesizer 113 of the codebook parameter generator 103. The codebook parameter generating means 103 searches the codebooks 104a to 104d for adaptive excitation vectors and noise excitation vectors using the speech synthesis means 113 in which filter coefficients based on the LSP parameters are set, so that the adaptive code A book parameter ia, an adaptive gain codebook parameter iga, a noise codebook parameter is, and a noise gain codebook parameter igs are generated.

[0009] The FEC 108 has an adaptive codebook parameter ia that significantly degrades reproduced voice quality when a transmission error occurs.
And an 11-bit information vector consisting of the specific bits of the adaptive gain codebook parameter iga and the specific bits of the extension parameter generated by the extension parameter generating means (not shown), the 4 bits by the (15,11) Hamming code method. Generate parity bits of The multiplexer 109 further includes a linear prediction parameter and an extension parameter for an n-th frame (n is an arbitrary integer),
By multiplexing each of the codebook parameters ia, iga, is, and igs of the first to fourth subframes of the n-th frame with the parity bits, a transmission block of the n-th frame is generated. To the transmission path 301.

FIG. 7 is a block diagram of a conventional speech decoding apparatus. 7 includes a multiplexer 209, an error correction unit 208, a subframe linear interpolation unit 102, a speech synthesis unit 203, an adaptive codebook 1
04a, adaptive gain codebook 104b, and random codebook 10
4c and a noise gain codebook 104d. In FIG. 7, the same components as those in FIG. 6 are denoted by the same reference numerals.

The multiplexer 207 receives the transmission block from the transmission path 301, separates the transmission block,
The LSP parameter is sent to the subframe linear interpolation means 102, the adaptive codebook parameter ia, the adaptive gain codebook parameter iga, the extension parameter and the parity bit are sent to the error correction means 208, and the noise codebook parameter is and the noise gain codebook are sent. The parameter igs is sent to the speech synthesis means 203.

The error correction means 208 includes an information vector including an adaptive codebook parameter ia, an adaptive gain codebook parameter iga, and specific bits of an extension parameter.
A transmission error of a transmission vector composed of a parity flag according to the (15, 11) Hamming code method for this information vector is checked according to the (15, 11) Hamming code method,
When a correctable error is detected, the error bit is corrected. When an uncorrectable error is detected, the adaptive codebook parameter ia and the adaptive gain codebook parameter i are corrected.
ga is corrected by substituting ia and iga of the first to fourth subframes of the previous frame (the (n-1) th frame), and replacing the extension parameter with the extension parameter of the previous frame. The corrected and inspected / corrected / corrected adaptive codebook parameter ia and adaptive gain codebook parameter iga are converted to speech synthesis means 203.
, And sends the extended / inspected / corrected / corrected extended parameters to extended function control means (not shown).

The subframe linear interpolation means 102 converts the LSP parameters in units of frames into filter coefficients in units of subframes, and
03 is set in the voice synthesis filter 113. Further, the speech synthesis means 203 receives the input codebook parameter i
a, iga, is, igs, and based on codebooks 104a-
An adaptive sound source vector and a noise sound source vector are searched from 104d, and a sound signal is synthesized by driving the sound synthesis filter 113 in which a filter coefficient based on the LSP parameter is set by the searched sound source vector.

[0014]

However, in the above-mentioned conventional speech coding / decoding apparatus, bits which are not protected by the error correction method exist in the information block.
If an error occurs in the unprotected bits, the quality of the reproduced sound may be small, but the reproduced sound may be difficult to hear. For example, when a bit error occurs in an unprotected LSP parameter, the reproduced sound may be turned upside down. Further, when a bit error occurs in the unprotected noise codebook parameter, the voice may be interrupted or an abnormal sound may occur at that portion.

SUMMARY OF THE INVENTION The present invention has been made to solve such a conventional problem. An error correction code generation process capable of enhancing the transmission error resistance of an information block without increasing the transmission bit rate as much as possible. It is an object to provide a method and an error correction processing method.

[0016]

In order to achieve the above object, an error correction code generation method according to the present invention provides an error correction code for generating an error correction code to be added to a transmitted information block for checking a transmission error. A method of generating a first error correction code for a first information vector of the information block according to a first error correction method, and a method of generating a first error correction code for the second information vector of the information block. Generating a second error-correcting code according to a second error-correcting method in which the ratio of the number of generated parity bits to the number of bits of the information vector is smaller than that of the first error-correcting method. .

Further, the error correction processing method of the present invention inspects a transmission error of a received transmission block including an information block and an error correction code, and corrects or corrects the information block based on the inspection result. A processing method for checking a transmission error of a transmission vector including a first information vector of the information block and a first error correction code in accordance with a first error correction method; A step of checking a transmission error of a transmission vector composed of a second error correction code according to a second error correction method, and a step of including a bit constituting the first information vector, and forming a bit constituting the second information vector. Correcting or correcting the data of the information block that does not include the data based on the inspection result by the first error correction technique; A step of correcting or correcting the data of the information block not including the bits configuring the information vector of the second information vector but including the bits configuring the second information vector, based on an inspection result by the second error correction technique; Correcting or correcting the data of the information block including the bits constituting the first and second information vectors based on the inspection result by the first error correction method and the inspection result by the second error correction method. And a procedure.

[0018]

FIG. 1 is a block diagram showing a speech coding apparatus according to an embodiment of the present invention. 1 includes a linear prediction analysis unit 101, a subframe linear interpolation unit 102, and a codebook parameter generation unit 103.
, Adaptive codebook 104a, adaptive gain codebook 104b
, Noise codebook 104c, and noise gain codebook 104d
And first error correction code generation means (1FEC) 105
And second error correction code generation means (2FEC) 106a
To 106 e and a multiplexer 107.
In FIG. 1, the same components as those in FIG. 6 are denoted by the same reference numerals.

The linear prediction analysis means 101 generates LSP parameters for each frame by performing linear prediction analysis on an input speech signal having a predetermined frame length (usually about 20 to 40 [ms]). Interpolation means 1
02 and the multiplexer 107. The input audio signal is digitized at a sampling frequency of 8 [kHz]. The subframe linear interpolation means 102 performs linear interpolation on the LSP parameters in frame units, so that the subframe unit is a subframe (a subframe is obtained by dividing a frame into four, and the subframe length is about 5 to 10 [ms]). ) To generate LSP parameters,
Further, the LSP parameters in subframe units are converted into filter coefficients, and the filter coefficients are set in the speech synthesis filter 113 of the codebook parameter generation means 103.

The codebook parameter generating means 103 has a search control means 111, a sound source synthesizing means 112, a speech synthesizing filter 113, and a distance calculating means 114.
The codebooks 104a to 104d are respectively searched by the -S (Analysis-by-Synthesis: synthesis) method, and the adaptive codebook parameter ia, the adaptive gain codebook parameter iga, the noise codebook parameter is, and the noise gain in subframe units are searched. A codebook parameter igs is generated, and these four types of codebook parameters are sent to the multiplexer 107, and the adaptive codebook parameter ia and the adaptive gain codebook parameter iga are sent to the 1FEC 105. The search control means 111 controls the search by the AbS method. The sound source synthesizing unit 112 synthesizes the adaptive excitation vector ea, the adaptive excitation vector gain ega, the noise excitation vector es, and the noise excitation vector gain egs retrieved from the codebooks 104a to 104d, respectively, to thereby obtain the speech synthesis filter 1
A sound source vector e for driving the sound source 13 is generated. In the speech synthesis filter 113, a filter coefficient based on the LSP parameter is set by the subframe linear interpolation means 102. The distance calculation means 114 includes a speech synthesis filter 113
Calculates the distortion (distance) of the synthesized speech due to the input speech. The AbS method described above uses codebooks 104a to 104a.
d, sound source synthesizing means 112, and voice synthesis filter 113
And a sound source vector (a combination of an adaptive sound source vector ea, an adaptive sound source vector gain ega, a noise sound source vector es, and a noise sound source vector gain egs) that minimizes the distance between the synthesized sound and the input sound by using a closed loop by the distance calculating means 114. This is a closed loop search method for searching, and search control means 1
Numeral 11 generates codebook parameters ia, iga, is, and igs indicating the adaptive excitation vector ea, the adaptive excitation vector gain ega, the noise excitation vector es, and the noise excitation vector gain egs that minimize the distance to the input speech signal.

Codebooks 104a to 104d store an excitation vector composed of an adaptive excitation vector expressing a periodic component (pitch) of a voice and a noise excitation vector expressing a change portion of the voice that cannot be expressed by the adaptive excitation vector. It is. Adaptive codebook 104a stores the adaptive excitation vector, adaptive gain codebook 104b stores the gain for the adaptive excitation vector, and noise codebook 104c stores the noise excitation vector. The gain codebook 104d stores the gain for the noise excitation vector.

1 FEC 105 is the same as FEC 108 in FIG. 6, and has a specific bit of adaptive codebook parameter ia and adaptive gain codebook parameter iga that significantly degrades reproduced voice quality when a transmission error occurs, and a specific bit of an extension parameter. The 11-bit first information vector consisting of
Generate parity bits for the bits. Note that the extension parameter is a parameter for designating a function other than audio reproduction generated by an extension parameter generation unit (not shown), and is not an audio parameter.

The 2FECs 106a to 106e are used for a second information vector including a bit of an audio parameter that does not significantly degrade the reproduced audio quality even if a transmission error occurs.
A 1-bit parity bit is generated by a bit parity coding method (this 1-bit parity bit is called a parity flag by distinguishing it from a 4-bit parity bit by a (15, 11) Hamming coding method). 2FEC106a
Generates a parity flag (LSP parity flag) for the second information vector composed of bits of the linear prediction parameter LSP in frame units. 2FEC106
b generates a parity flag (adaptive codebook parity flag) for the second information vector composed of bits of the adaptive codebook parameter ia in subframe units. 2FE
C106c is a parity flag (adaptive gain codebook parity flag) for the second information vector composed of bits of the adaptive gain codebook parameter iga in subframe units.
Generate The 2FEC 106d generates a parity flag for the second information vector including the bits of the noise codebook parameter igs in subframe units. 2FE
C106e is a parity flag (noise gain codebook parity flag) for the second information vector including the bits of the noise gain codebook parameter igs for each subframe.
Generate

The multiplexer 107 includes an information block of the n-th frame including an LSP parameter of the n-th frame, codebook parameters of the first to fourth sub-frames of the n-th frame, and an extension parameter of the n-th frame;
Parity bits for the first information vector of the n-th frame and 4 of the first to fourth sub-frames of the n-th frame
The transmission block of the n-th frame is generated by multiplexing the second information vector with the error correction code composed of each parity flag, and the transmission block is transmitted to the transmission path 301.

Next, the operation of the speech coding apparatus shown in FIG. 1 will be described. Since the process of generating the voice parameters is based on a known technique, the description thereof will be omitted. Assuming that the order of the LSP parameters generated by the linear prediction analysis unit 101 in units of frames is 10 (0th to 9th), 1
The number of bits of the LSP parameter for the frame is 34. Further, in each codebook parameter generated in subframe units, the number of bits of adaptive gain codebook parameter iga and the number of bits of noise gain codebook parameter iga are each 5, and the number of bits of noise codebook parameter is is 9; The number of bits of the book parameter ia is 8 for odd subframes (first and third subframes) and 6 for even subframes (second and fourth subframes). In the even-numbered subframe, the number of bits of the adaptive codebook parameter is reduced by using the difference from the adaptive codebook parameter (8 bits) of the immediately preceding odd-numbered subframe as the adaptive codebook parameter. The rate is reduced.

FIG. 2 is a flowchart of an error correction code generation process in the speech coding apparatus of FIG. First, as shown in step S1, in the 1FEC 105, the n-th
For the first information vector of the information block of the frame,
(15, 11) Four parity bits are generated by the Hamming code method. Assuming that 11 bits forming the 11-bit first information vector are x0 to x10, x0
Is M of the adaptive codebook parameter ia of the first subframe.
SB (Most Significant Bit), x1
Is 6 of the adaptive codebook parameter ia of the first subframe.
The bit, x2 is the fifth hit of the adaptive codebook parameter ia of the first subframe, x3 is the MSB of the adaptive gain codebook parameter iga of the first subframe, and x4 is the adaptive gain codebook of the second subframe. M of the parameter iga
SB, x5 is the MSB of the adaptive codebook parameter ia of the third subframe, x6 is the sixth bit of the adaptive codebook parameter ia of the third subframe, and x7 is the adaptive codebook parameter ia of the third subframe. 5th bit, x8
Is the adaptive gain codebook parameter ig of the third subframe
The MSB of a, x9 is the MSB of the adaptive gain codebook parameter iga of the fourth subframe, and x10 is the MSB of the extension parameter.

Let the first information vector be X = (x0, x1,
, X10), and the transmission vector of this first information vector X by the (15, 11) Hamming code method is W
= (W0, w1,..., W10, w11,..., W14)
(Where w0 = x0, w1 = x1,..., W10 = x1
1 and w11 to w14 are parity bits), the transmission vector W is generated using the Hamming code generation matrix G by the operation of W = X · G (1). Here, the Hamming code generation matrix G is

(Equation 1) It is. FIG. 3 is a diagram for explaining a process of generating parity bits by the (15, 11) Hamming code method shown in Expression (1). As shown in FIG. 3, parity bits w11 to w
14 is generated by the operation of each w11 = x0 + x1 + x3 + x4 + x6 + x8 + x10 (2) w12 = x0 + x2 + x3 + x5 + x6 + x9 + x10 (3) w13 = x1 + x2 + x3 + x7 + x8 + x9 + x10 (4) w14 = x4 + x5 + x6 + x7 + x8 + x9 + x10 (5). However, (2) to (5)
“+” In the expression indicates an exclusive OR (XOR) operation. That is, in FIG. 3, each of the parity bits is obtained by performing an exclusive OR operation on the bits of the first information vector with a band line.

Here, a brief description will be given of a transmission error check process by the (15, 11) Hamming code method. The transmission vector W generated on the audio encoding side and transmitted to the transmission path is referred to as a transmission transmission vector, and the transmission vector received from the transmission path on the audio decoding side is referred to as a reception transmission vector. A transmission error vector E is added to W in the transmission path, and the received transmission vector is Y = W + E
It is assumed that At this time, the product S = Y · H ^T of a transposed matrix H ^T of the received transmission vector Y and a Hamming code parity check matrix H (6) referred to as syndromes for Y. Here, the Hamming code check matrix H is

(Equation 2) It is. Since W.H ^T = O (O indicates a zero matrix or a zero vector), the syndrome S is expressed as follows: S = Y · H ^T = (W + E) · H ^T = W · H ^T + E · H ^T = E · H ^T (7), which is determined only by the transmission error vector E. S =
If O, there is no transmission error (E = O), indicating that the reception transmission vector Y matches the transmission transmission vector W. When the syndrome S matches the i-th column (i is an arbitrary integer from 1 to 15) of the Hamming code check matrix H,
Only the ith component of the transmission error vector E is 1, indicating that a single error has occurred only in the ith component of the received transmission vector Y. Therefore, if the i-th component of the reception transmission vector Y is corrected, the transmission transmission vector W can be obtained. When the syndrome S is not the zero matrix O and does not match any column of the Hamming code check matrix H, it indicates that a transmission error has occurred in a plurality of bits.

Next, as shown in steps S2 to S9 of FIG. 2, in the 2FECs 106a to 106e,
A parity flag is generated for the second information vector of the information block of the n-th frame by a 1-bit parity coding method. The LSP parameter, the respective adaptive codebook parameters ia of the first to fourth subframes, the respective adaptive gain codebook parameters iga, the respective noise codebook parameters is, and the respective second noise gain codebook parameters igs are used to individually and individually determine the second.
, The information block of the n-th frame contains a total of 17 second information vectors.

The process of generating a parity flag by the 1-bit parity coding method is performed by using a 1-bit information vector composed of q bits x (1) to x (q) (q is a positive integer).
Generate a bit parity flag w (q + 1) (q +
1-bit transmission vector). This one-bit parity coding method has no error correction capability, and can detect an error only when a transmission error occurs in an odd number of bits of a transmission vector.

In the process of checking a transmission error by the 1-bit parity code method, the bits of the received transmission vector corresponding to the bits x (1) to x (q) are set to y (1) to y (y).
(1) If the received transmission vector corresponding to the parity flag w (q + 1) is y (q + 1), PC = y (1) + y (2) +... Y (q) + y (q + 1) (14) If the check flag PC is 0, there is no transmission error and 1
If so, it is determined that a transmission error has occurred.

First, the process of generating the LSP parity flag shown in step S2 will be described. Here, 34
Among the LSP parameters of bits, the 0th to 9th M
The SB 10 bits are bits constituting the second information vector, and the other 10 bits are bits that are very unlikely to cause an auditory problem even if a transmission error occurs. 2 is excluded from the information vector. If the information bit corresponding to the MSB of the p-th LSP parameter of the n-th frame (p is any integer from 0 to 9) is x [LSP (n, p, MSB)], the LSP parity of the n-th frame The flag w [PLSP (n)] is 2FEC1
In 06a, w [PLSP (n)] = x [LSP (n, 0, MSB)] + x [LSP (n, 1, MSB)] + x [LSP (n, 2, MSB)] + x [LSP (n, 3, MSB)] + x [LSP (n, 4, MSB)] + x [LSP (n, 5, MSB)] + x [LSP (n, 6, MSB)] + x [LSP (n, 7, MSB)] + x [LSP (n, 8, MSB)] + x [LSP (n, 9, MSB)] (8) Note that a 1-bit LSP parity bit may be generated for the second information vector composed of all the bits of the LSP parameter.

Next, the process of generating each codebook parity flag will be described. As shown in step S4, the k1th bit (k1 is 1 to 8 when m is an odd number, even number) of the adaptive codebook parameter of the mth subframe of the nth frame (m is any integer from 1 to 4). X = [ia (m, k1)] corresponding to an information bit corresponding to an arbitrary integer from 1 to 6
Then, the adaptive codebook parity flag w [Pia (m)] of the m-th subframe becomes m in the 2FEC 106b.
Is odd, w [Pia (m)] = x [ia (m, 1)] + x [ia (m, 2)] + x [ia (m, 3)] + x [ia (m, 4)] + x [Ia (m, 5)] + x [ia (m, 6)] + x [ia (m, 7)] + x [ia (m, 8)] (9), and when m is even, w [ Pia (m)] = x [ia (m, 1)] + x [ia (m, 2)] + x [ia (m, 3)] + x [ia (m, 4)] + x [ia (m, 5) ] + X [ia (m, 6)] (10)

As shown in step S5, an information bit corresponding to the k2th bit (k2 is an arbitrary integer from 1 to 5) of the adaptive gain codebook parameter of the mth subframe is x [iga (m , k2)], the adaptive gain codebook parity flag w [Piga (m)] of the m-th subframe.
In the 2FEC 105c, w [Piga (m)] = x [iga (m, 1)] + x [iga (m, 2)] + x [iga (m, 3)] + x [iga (m, 4)] + x [Iga (m, 5)] Generated by (11).

As shown in step S6, the information bit corresponding to the k3th bit (k3 is an arbitrary integer from 1 to 9) of the noise codebook parameter of the m-th subframe is x [is (m, k3)], the noise codebook parity flag w [Pis (m)] of the m-th subframe is 2FEC
At 105d, w [Pis (m)] = x [is (m, 1)] + x [is (m, 2)] + x [is (m.3)] + x [is (m, 4)] + x [is (m, 5)] + x [is (m, 6)] + x [is (m, 7)] + x [is (m, 8)] + x [is (m, 9)] (12)

Further, as shown in step S7, if the information bit corresponding to the k2th bit of the noise gain codebook parameter of the mth subframe is x [igs (m, k2)], the mth subframe In the 2FEC 105e, the noise gain codebook parity flag w [Pigs (m)] = w [Pigs (m)] = x [igs (m, 1)] + x [igs (m, 2)] + x [igs (m , 3)] + x [igs (m, 4)] + x [igs (m, 5)] (13) However, “+” in the above equations (8) to (13) indicates an XOR operation. Note that if there is a bit in the codebook parameter that has a very low possibility of causing a hearing problem even if a transmission error occurs, the 1-bit parity coding method is applied to all bits of the codebook parameter. Instead, the 1-bit parity coding method may be applied to bits excluding bits that are unlikely to cause the above-described audibility problem.

After the LSP parity flag is generated in step S2, the subframe variable m is set to 1 in step S3, each codebook parity flag of the first subframe is generated in steps S4 to S7, and then m> in step S8. If it is 3, the process proceeds to step S9. In step S9, m is incremented to 2 and the process returns to step S4.
7, each codebook parity flag of the second sub-frame is generated, and thereafter, each codebook parity flag of the third and fourth sub-frames is generated in a loop of steps S4 to S9, and the error code generation process ends.

As described above, the LSP parity flag is generated for each frame, and the adaptive codebook parity flag, adaptive gain codebook parity flag, noise codebook parity flag, and noise gain codebook parity flag are respectively generated for each subframe. As a result, the transmission block includes a parity flag of a total of 17 bits.
Note that the generation of the parity bit by the (15, 11) Hamming code method and the generation of the parity flag by the 1-bit parity code method may be performed at the same time. May be generated.

The (15,
11) An n-th parity bit including a 4-bit parity bit by the Hamming code method and a parity flag by the 1-bit parity code method generated by the 2FECs 105a to 105e.
The error correction code of the frame includes the LSP parameter of the n-th frame generated by the linear prediction analysis unit 101 and the codebook parameters of the first to fourth subframes of the n-th frame generated by the codebook parameter generation unit 103. Is transmitted to the multiplexer 107 together with the information block including the extension parameters of the n-th frame. The multiplexer 107 generates a transmission block of the n-th frame by multiplexing the information block and the error correction code, and
To the audio decoding device via the.

[0040]

[Table 1]

[Table 2] (Table 1) shows the bit configuration of a transmission block in the speech coding apparatus of FIG. 1, and (Table 2) shows the bit configuration of a transmission block in the conventional speech coding apparatus. As shown in (Table 1) and (Table 2), an error correction code in a transmission block of a conventional speech coding apparatus (hereinafter, simply referred to as a conventional transmission block) uses a parity (15,11) Hamming code. The error correction code in the transmission block (hereinafter, simply referred to as the transmission block of the present invention) of the speech encoding device of the present invention is only 4 bits.
(15, 11) A total of 21 bits obtained by adding a 17-bit parity flag according to a 1-bit parity code to a 4-bit parity bit according to a Hamming code, and
The information block in the conventional transmission block and the information block in the transmission block of the present invention are both 34
LSP parameter of 28 bits, adaptive codebook parameter of 28 bits, adaptive gain codebook parameter and noise gain codebook parameter of 20 bits each, noise codebook parameter of 36 bits, and extension parameter of 2 bits, totaling 140 Is a bit. Therefore, the conventional transmission block has 144 bits, and the transmission block of the present invention has 161 bits, which is 17 bits larger than the conventional transmission block. Assuming that the frame length is 30 [ms], the transmission bit rate is 4.8 [kbps] in the conventional transmission block, and 11 bits are protected by the error correction code in the 140-bit information block. On the other hand, in the transmission block of the present invention, the transmission bit rate is 5.3 [kbps], and in the 140-bit information block, all 140 bits of the information block can be protected by the error correction code (however, It is assumed that the second information vector of the LSP parity flag is composed of all bits of the LSP parameter.) In a conventional speech decoding apparatus, when all 140 bits of an information block are protected by the (15, 11) Hamming code method, the generated parity bits are 52 (= 4 × 1).
3) bits, the transmission block becomes 192 bits, and the transmission bit rate increases to 6.4 [kbps].

FIG. 4 is a block diagram showing a speech decoding apparatus according to an embodiment of the present invention. The audio decoding device in FIG.
Multiplexer 207 and first error correction means 105
And second error correction means 206a, 206d, 206e
, Subframe linear interpolation means 102, speech synthesis means 203, adaptive codebook 104a, adaptive gain codebook 10
4b, the noise codebook 104c, and the noise gain codebook 104
d. In FIG. 4, the same components as those in FIGS. 1 and 7 are denoted by the same reference numerals.

The multiplexer 207 converts the transmission block of the n-th frame received from the transmission path 301 into an information block and an error correction code, that is, LSP parameters, codebook parameters ia, iga, is, igs, extension parameters, (15, 11) Parity bits by the Hamming code method and respective parity flags are separated, and the first error correction means 205 sends the adaptive codebook parameter ia and the adaptive gain codebook parameter iga of the first to fourth subframes.
, An extension parameter, and a parity bit, and send an LSP parameter LSP and an LSP parity flag to the second error correction means 206a.
06d, the noise codebook parameter is of the first to fourth subframes and the noise codebook parity flag of the first to fourth subframes are sent to the second error correction means 206e.
The noise gain codebook parameter igs of the fourth subframe and the noise codebook parity flags of the first to fourth subframes are sent.

The first error correction means 205 outputs (15, 1
1) According to the Hamming code method, the first codec parameter ia of the first and third subframes, the adaptive gain codebook parameter iga of the first to fourth subframes, and the first 11 bits that are specified as the extension parameters. The transmission error of the transmission vector including the information vector and the 4-bit parity flag is checked, and based on the check result, the adaptive codebook parameters ia of the first and third subframes,
Adaptive gain codebook parameter iga of the fourth subframe,
And the extension parameters are corrected or replaced, and only the extension parameters are sent to the extension function control means (not shown).
The error correction processing using the (15, 11) Hamming code method corrects the corresponding bit when an error that can be corrected is detected, and when an uncorrectable error is detected, only the extension parameter is used. Is replaced by the data of the previous frame. Next, according to the 1-bit parity coding method, each transmission vector including each adaptive codebook parameter ia (each second information vector) of the first to fourth subframes and each corresponding adaptive codebook parity flag,
And a transmission error of each transmission vector comprising each adaptive gain codebook parameter iga (each second information vector) of each of the first to fourth subframes and each corresponding adaptive gain codebook parity flag, and an error is detected. Sometimes (the 1-bit parity code method has no error correction capability), the correction is performed by replacing the codebook parameter with the codebook parameter of the previous subframe, and the adaptive code after the error correction processing by the 1-bit parity code method is completed. The book parameter ia and the adaptive gain codebook parameter iga are converted to speech synthesis means 20.
Send to 3.

The second error correction means 206a checks the second information vector consisting of the MSB of the LSP parameter and the transmission vector consisting of the LSP parity flag according to the 1-bit parity coding method, and when an error is detected,
This LSP parameter is replaced with the LSP parameter of the previous frame, and the LSP parameter after the error correction processing is sent to the subframe linear interpolation means 102. Also, the second
The error correction means 206d checks the transmission vector composed of the noise codebook parameter is and the corresponding noise codebook parity flag according to the 1-bit parity code method, and when an error is detected, this noise codebook parameter is The noise codebook parameter is replaced with the noise codebook parameter of the frame, and the noise codebook parameter is subjected to the error correction processing is sent to the speech synthesizer 203. Also, the second error correction means 20
6e checks the transmission code composed of the noise gain codebook parameter igs and the noise gain codebook parity flag according to the 1-bit parity code method, and when an error is detected,
This noise gain codebook parameter igs is replaced by the noise gain codebook parameter of the previous subframe, and the noise codebook parameter is that has been subjected to the error correction processing is replaced by the speech synthesis means 2.
Send to 03.

The speech synthesizing means 203
1, sound source synthesizing means 112, and voice synthesizing filter 113
Adaptive excitation vector ea and adaptive excitation vector gain eg respectively searched from codebooks 104a to 104d by search control means 211 according to codebook parameters input from second error correction means 206b to 206e.
a, noise source vector es, noise source vector gain eg
s is synthesized by the sound source synthesizing means 112 to generate a sound source vector e.
The audio signal is reproduced by driving the audio synthesis filter 113, in which the filter coefficient based on the SP parameter is set, with the sound source vector e.

Next, the operation of the speech decoding apparatus shown in FIG. 4 will be described. Note that the process of reproducing the audio signal from the error-corrected audio parameter is based on a known technique, and a description thereof will be omitted. The transmission block of the n-th frame transmitted from the speech coding apparatus of FIG. 1 to the transmission path 301 is received by the multiplexer 209 and separated into an information block and an error correction code. Further, this information block includes LSP parameters, codebook parameters ia, iga, is, igs of the first to fourth subframes,
It is separated into an extension parameter, a parity bit, and each parity flag. The adaptive codebook parameter ia, the adaptive gain codebook parameter iga, the extension parameter, the parity bit, the adaptive codebook parity flag, and the adaptive gain codebook parity bit are sent to the first error correction means 205, where the LSP parameter and the LSP parity The flag is sent to the second error correction means 206a, the noise codebook parameter is and the noise codebook parity flag are sent to the second error correction means 206d, and the noise gain codebook parameter igs and the noise codebook parity flag are sent to the second error correction means 206d. 2 error correction means 206e.

FIG. 5 is a flowchart of the error correction process in the speech decoding apparatus of FIG. First, step S1
As shown in 1 to S15, the first error correction means 205 includes a first information vector including specific bits of the adaptive codebook parameter ia, the adaptive gain codebook parameter iga, and the parameter for extension, and its parity bit. An error correction process is performed on the transmission vector by the (15, 11) Hamming code method. In step S11,
The transmission error of the above-described transmission vector received by the multiplexer 207 is represented by the Hamming check matrix H shown in (Equation 2).
, And if there is no transmission error in step S12, the error correction process ends. If a transmission error has been detected, the process proceeds to step S13. Step S
In step 13, if the transmission error is correctable, the corresponding bit is corrected in step S14. If the transmission error is not correctable, in step S15,
The correction is performed by replacing the extension parameter of the nth frame with the extension parameter of the previous frame ((n-1) th frame). The first error correction means 205 outputs (1
5, 11) The extension parameters that have been subjected to the error correction processing (inspection and error correction or error correction) by the Hamming code method are sent to an extension function control unit (not shown). Step S
Even if the error detected in step 13 cannot be corrected, in step 15, the adaptive codebook parameter ia and the adaptive gain codebook parameter iga are not corrected. After that, the codebook parameters ia and iga are checked by the 1-bit parity coding method, and then whether or not to correct them is determined.

Next, as shown in steps S16 to S33, the first error correction means 205 and the second error correction means 206a, 206d, and 206e execute each second information vector of the information block of the n-th frame. An error correction process based on the 1-bit parity coding method is performed on each transmission vector including the corresponding parity flag. However, since the 1-bit parity coding method has no error correction capability, the above error correction processing is an error check and an error correction. In the error check by the 1-bit parity code method, as described above, a check flag PC is generated, and when this PC is 1, it is determined that a transmission error has occurred. Error correction when an error is detected is to replace a parameter corresponding to a second information vector of a transmission vector in which an error is detected with a parameter of a previous frame or a previous subframe.

First, in the second error correction means 206a, as shown in steps S16 and S17, the LSP
Transmission vector (second information vector and LSP parity flag configured by MSB of LSP parameter)
Is checked, and when an error is detected, the error is corrected by replacing the LSP parameter of the n-th frame with the (n-1) -th LSP parameter, as shown in step S18. The received bit corresponding to the transmitted bit x [LSP (n, p, MSB)] of the LSP parameter of the n-th frame is represented by y [LSP (n, p, MSB)], and the transmitted LSP parity flag w [ PLSP (n)] received LSP
Assuming that the parity flag is y [PLSP (n)], the check flag PC is: PC = y [LSP (n, 0, MSB)] + y [LSP (n, 1, MSB)] + y [LSP (n, 2, MSB)] + y [LSP (n, 3, MSB)] + y [LSP (n, 4, MSB)] + y [LSP (n, 5, MSB)] + y [LSP (n, 6, MSB)] + y [LSP (n, 7, MSB)] + x [LSP (n, 8, MSB)] + x [LSP (n, 9, MSB)] + w [PLSP (n)] (15) When PC = 1, the bit y [LSP (n, p, MSB)] of the n-th frame is replaced with the bit y [LSP (n-1, p, MSB)] of the (n-1) -th frame. To compensate for this. The LSP parameters that have been subjected to the error correction processing by the 1-bit parity coding method in the second error correction means 206a are sent to the subframe linear interpolation means 102.

Next, the first error correction means 205 checks the transmission error of the received adaptive codebook transmission vector (adaptive codebook parameter and adaptive codebook parity flag) as shown in steps S20 and S21. ,
If an error is detected, a correction process is performed on the adaptive codebook parameters as shown in step S22. In the adaptive codebook parameter of the m-th sub-frame of the n-th frame, the received bit corresponding to the transmitted bit x [ia (m, k1)] is y [ia (m, k1)] and the transmitted adaptation is If the received adaptive codebook parity flag corresponding to the codebook parity flag w [Pia (m)] is y [Pia (m)], the check flag P
C is PC = y [ia (m, 1)] + y [ia (m, 2)] + y [ia (m, 3)] + y [ia (m, 4)] + y [ia when m is an odd number (m, 5)] + y [ia (m, 6)] + y [ia (m, 7)] + y [ia (m, 8)] + y [Pia (m)] (16) where m is an even number PC = y [ia (m, 1)] + y [ia (m, 2)] + y [ia (m, 3)] + y [ia (m, 4)] + y [ia (m, 5)] + Y [ia (m, 6)] + y [Pia (m)] (17) When PC = 1, the bit y [ia (m, k1)] of the m-th sub-frame is replaced by the bit y [ia (m-1, k1)] of the (m-1) -th sub-frame. Correct (however, when m = 1, replace with the bit of the fourth sub-frame of the previous frame). As described above, the adaptive codebook parameter of the second or fourth subframe is difference data from the adaptive codebook parameter of the immediately preceding first or third subframe. Alternatively, the conversion is performed after converting the adaptive codebook parameter of the fourth subframe into absolute data.

Next, in the first error correction means 205, as shown in steps S23 to S25, the received adaptive gain codebook transmission vector (adaptive gain codebook parameter and adaptive gain codebook parity flag) has one bit. An error correction process is performed by a parity code method. The received bit corresponding to the transmitted bit x [iga (m, k2)] of the adaptive gain codebook parameter of the m-th subframe of the n-th frame is represented by y [iga (m, k2)], and the transmitted adaptation is transmitted. If the received adaptive gain codebook parity flag corresponding to the gain codebook parity flag w [Piga (m)] is y [Piga (m)],
The inspection flag PC is calculated as follows: PC = y [iga (m, 1)] + y [iga (m, 2)] + y [iga (m, 3)] + y [iga (m, 4)] + y [iga (m, 5 )] + Y [Piga (m)] (18) When PC = 1, the bit y [ig
a (m, k2)] is replaced by the bit y [iga (m-1, k
2) Replace with].

Next, in the second error correction means 206d, as shown in steps S26 to S28, the received noise codebook transmission vector (the noise codebook parameter and the noise codebook parity flag) is subjected to the one-bit parity code method. Error correction processing. The received bit corresponding to the transmitted information bit x [is (m, k3)] of the noise codebook parameter of the m-th subframe of the n-th frame is represented by y [is
(m, k3)] and the transmitted random codebook parity flag w [Pi
If the received noise codebook parity flag corresponding to s (m)] is y [Pis (m)], the check flag PC is: PC = y [is (m, 1)] + y [is (m, 2 )] + Y [is (m.3)] + y [is (m, 4)] + y [is (m, 5)] + y [is (m, 6)] + y [is (m, 7)] + y [is (m, 8)] + y [is (m, 9)] (19) When PC = 1, the bit y [is
(m, k3)] with the bit y [is (m-1, k3)] of the previous subframe.
Replace with

Next, in the second error correction means 206e, as shown in steps S29 to S31, 1 is added to the received noise gain codebook transmission vector (noise gain codebook parameter and received noise gain codebook parity flag). An error correction process using a bit parity code method is performed. The received bit corresponding to the transmitted bit x [igs (m, k2)] of the noise gain codebook parameter of the mth subframe of the nth frame is represented by y [igs (m, k2)], and the transmitted adaptation Assuming that the received adaptive gain codebook parity flag corresponding to the gain codebook parity flag w [Pigs (m)] is y [Pigs (m)], the check flag PC is: PC = y [igs (m, 1) ] + Y [igs (m, 2)] + y [igs (m, 3)] + y [igs (m, 4)] + y [igs (m, 5)] + y [Pigs (m)] (20) You. When PC = 1, the bit y [ig
s (m, k2)] is replaced with information bits y [igs (m−
1, k2)]. The second error correction means 206
The adaptive codebook parameter ia, adaptive gain codebook parameter iga, and noise codebook parameter i which have been subjected to error correction by the 1-bit parity code method in b to 206e.
s and the noise gain codebook parameter igs are sent to the search control means 211 of the speech synthesis means 203.

When the error correction processing of the LSP parameter is completed in steps S16 to S18, the subframe variable m is set to 1 in step S19, and error correction processing is performed on each codebook parameter of the first subframe in steps S20 to S31. Next, if m> 3 in step S32, step S3
In step S33, m is incremented to 2 in step S33, and the process returns to step S20. In steps S20 to S31, error correction processing is performed on each codebook parameter of the second subframe. An error correction process is performed on each codebook parameter of the third and fourth subframes, and the error correction process ends.

When the LSP parameters are corrected continuously for N frames (N is an integer of 2 or more), the same LSP parameters are used continuously for N frames, and the sound quality deteriorates in terms of hearing. Also, M subframes (M
Is an integer of 2 or more). Similarly, when codebook parameters are continuously corrected, sound quality deteriorates in terms of hearing. Therefore, LS
When the P parameter is corrected for three consecutive frames, and when the codebook parameter is corrected for four subframes, the level of the reproduced audio signal is attenuated stepwise by waveform processing. Deterioration of hearing characteristics.

As described above, according to the embodiment of the present invention,
For the first information vector of the information block to be transmitted, (1
(5, 11) Parity bits are generated by the Hamming code method, and bits (15, 11) that are not protected by the Hamming code method and that have relatively little deterioration in reproduced voice quality even when a transmission error occurs, , 11) A parity flag is generated by a 1-bit parity code method in which the ratio of the number of parity bits to the number of bits of an information vector is smaller than that of the Hamming code method (resulting in a lower error correction capability), and an information block, a parity bit, and a parity flag are generated. By transmitting the transmission block consisting of the following, the number of bits of the information block protected by the error correction code can be increased by only slightly increasing the transmission bit rate, and the transmission error resistance can be increased. So you can
It is possible to suppress deterioration of the audibility characteristic due to a transmission error of a bit that is not protected.

In the above embodiment, two types of error correction methods, (15,11) Hamming code method and 1-bit parity code method, are used for information blocks.
The present invention is characterized in that two types of error correction methods are used for one information block in which the ratio of the number of parity bits to the number of bits of an information vector is different (resulting in different error correction capabilities). , Error correction techniques
The present invention is not limited to the (15, 11) Hamming code method and the 1-bit parity code method.

In the above embodiment, both (15, 11) Hamming code and 1-bit parity code are applied to specific bits of the adaptive codebook parameter and the adaptive gain codebook parameter. , The adaptive codebook parameter and the adaptive gain codebook parameter (15,
11) The 1-bit parity coding method may be applied only to bits to which the Hamming coding method is not applied.

[0059]

According to the present invention, as described above, the first information vector of the information block to be transmitted is
Generates a first error correction code according to the first error correction method, and generates a second information vector that is not protected by the first error correction method. 2nd bit ratio is small
Generating a second error correction code by the error correction method of
By transmitting the transmission block consisting of the information block and the first and second error correction codes, the bit of the information block protected by the error correction code can be far increased by slightly increasing the transmission bit rate. And the transmission error resistance can be increased, so that there is an effect that deterioration of audibility characteristics due to transmission errors of bits not protected can be suppressed.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech encoding device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing an error correction code generation process in the speech encoding device according to the embodiment of the present invention.

FIG. 3 is a diagram illustrating a process of generating parity bits by the (15, 11) Hamming code method.

FIG. 4 is a block diagram illustrating a configuration of a speech decoding device according to an embodiment of the present invention.

FIG. 5 is a flowchart showing an error correction process in the speech decoding device according to the embodiment of the present invention.

FIG. 6 is a block diagram illustrating a configuration of a conventional speech encoding device.

FIG. 7 is a block diagram illustrating a configuration of a conventional speech decoding device.

[Explanation of symbols]

101 linear prediction analysis means, 102 subframe linear interpolation means, 103 codebook parameter generation means, 1
04 codebook, 105 first error correction code generation means (1FEC), 106 second error correction code generation means (2FEC), 107,207 multiplexer,
203 speech synthesis means, 205 first error correction means, 206 second error correction means, 301 transmission path.

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Keiichi Chihara, Inventor 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (72) Makoto Morito 1-7-112 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (72) Inventor Takefumi Nakamura 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (72) Inventor Toshikazu Nakajima 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric (72) Inventor Takashi Yamagami 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd.

Claims (16)

[Claims] 1. An error correction code generation method for generating an error correction code to be added to a transmitted information block for checking a transmission error, wherein a first error vector is added to a first information vector of the information block. Generating a first error correction code in accordance with a correction method; and for the second information vector of the information block, the ratio of the number of generated parity bits to the number of information vector bits is smaller than that of the first error correction method. Generating a second error correction code according to a second error correction method. 2. The first error correction method is a method capable of correcting at least a single-bit transmission error of the first information vector, and the second error correction method is at least the first error correction method. 2. The method according to claim 1, wherein the method is capable of detecting a single-bit transmission error of the second information vector. 3. The method according to claim 1, wherein the first information is added to the information block.
And the total number of bits of the second error correction code is
And all bits constituting the second information vector,
2. An error correction code generated according to the first error correction method, the error correction code being less than the total number of bits of the error correction code added to the information block.
The error correction code generation method described in the above. 4. The method according to claim 1, wherein the first error correction code is based on a Hamming code, and the second error correction code is based on a one-bit parity code. The method for generating an error correction code according to any one of the above. 5. An error correction processing method for inspecting a transmission error of a received transmission block including an information block and an error correction code, and correcting or correcting the information block based on a result of the inspection. Checking a transmission error of a transmission vector composed of one information vector and a first error correction code according to a first error correction method; and transmitting a second information vector of the information block and a second error correction code. Checking the transmission error of the vector in accordance with a second error correction method, including the bits constituting the first information vector, and excluding the bits constituting the second information vector, A step of correcting or correcting based on an inspection result by the first error correction method, and a step of forming a bit constituting the first information vector. And correcting or correcting the data of the information block including the bits constituting the second information vector based on the inspection result by the second error correction technique, and Correcting the data of the information block including the bits constituting the information vector based on the inspection result by the first error correction method and the inspection result by the second error correction method. Error correction processing method. 6. The error correction processing method according to claim 5, wherein the correction of the data is performed by replacing the data with another data correlated with the data. 7. The first error correction method is a method capable of correcting at least a single-bit transmission error of the first information vector, and the second error correction method is at least the first error correction method. 6. The error correction processing method according to claim 5, wherein the method is capable of detecting a single-bit transmission error of the second information vector. 8. The method according to claim 5, wherein the first error correction method is a Hamming code method, and the second error correction method is a one-bit parity code method. Error correction processing method described. 9. Parameter generation means for generating a plurality of voice parameters based on an input voice signal, and error correction code generation for generating an error correction code for checking a transmission error of an information block including the voice parameters. Means, and a multiplexer for transmitting a transmission block composed of the information block and the error correction code to a voice decoding side, wherein the error correction code generation means includes: a first information vector of the information block; Means for generating a first error correction code in accordance with an error correction method; and a ratio of the number of generated parity bits to the number of information vector bits for the second information vector of the information block is higher than that of the first error correction method. Means for generating a second error correction code according to a small second error correction technique. Location. 10. The bit of the audio parameter included in the first information vector is a bit that significantly deteriorates the sound quality of a reproduced audio when a transmission error occurs, and is not included in the first information vector. 10. The speech encoding apparatus according to claim 9, wherein the bits of the speech parameter included in the second information vector are bits that cause little deterioration in the sound quality of the reproduced speech even if a transmission error occurs. 11. The speech parameter includes: a linear prediction parameter that specifies a filter coefficient of a speech synthesis filter; and a plurality of codebook parameters that specify an excitation vector that drives the speech synthesis filter. An adaptive codebook parameter specifying an adaptive excitation vector, an adaptive gain codebook parameter specifying an adaptive excitation vector gain, a noise codebook parameter specifying a noise excitation vector, and a noise gain code specifying a noise excitation vector gain Book parameter, the first information vector is an information vector including bits of the adaptive codebook parameter and the adaptive gain codebook parameter, and the second information vector is configured by bits of the linear prediction parameter. Information vector to be obtained, and the adaptive codebook parameter An information vector composed of bits of the adaptive gain codebook parameter, an information vector composed of bits of the noise codebook parameter, and an information vector composed of bits of the noise codebook parameter. 10. The speech encoding apparatus according to claim 9, wherein the information vector is an information vector to be encoded. 12. The method according to claim 9, wherein the first error correction code is based on a Hamming code method, and the second error correction code is based on a one-bit parity code method. The speech encoding device according to any one of the above. 13. A multiplexer for receiving a transmission block composed of an information block including a plurality of speech parameters and an error correction code from a speech code side, a transmission error of the transmission block is inspected, and the information block is checked based on a result of the inspection. Error correcting means that corrects or corrects, and voice synthesizing means for synthesizing a voice signal based on the voice parameter that has been subjected to the error correction processing by the error correcting means, wherein the error correcting means comprises: Means for checking a transmission error of a transmission vector composed of one information vector and a first error correction code in accordance with a first error correction method; and transmission of a second information vector of the information block and a second error correction code. Means for checking a transmission error of a vector according to a second error correction method; Means for correcting or correcting a speech parameter including a bit and not including a bit constituting the second information vector based on a check result by the first error correction technique; and Means for correcting or correcting a speech parameter not including a bit to be performed but including a bit constituting the second information vector based on an inspection result by the second error correction method; and configuring the first information vector. Means for correcting or correcting a speech parameter including a bit to be changed and a bit constituting the second information vector based on a check result by the first error correction technique and a check result by the second error correction technique. An audio decoding device comprising: 14. The bit of the voice parameter included in the first information vector is a bit that significantly deteriorates the sound quality of a reproduced voice when a transmission error occurs, and is not included in the first information vector. 14. The speech decoding apparatus according to claim 13, wherein the bits of the speech parameter included in the second information vector are bits that cause little deterioration in sound quality of reproduced speech even when a transmission error occurs. 15. The speech parameter includes: a linear prediction parameter for designating a filter coefficient of a speech synthesis filter; and a plurality of codebook parameters for designating an excitation vector for driving the speech synthesis filter. An adaptive codebook parameter specifying an adaptive excitation vector, an adaptive gain codebook parameter specifying an adaptive excitation vector gain, a noise codebook parameter specifying a noise excitation vector, and a noise gain code specifying a noise excitation vector gain Book parameter, the first information vector is an information vector including bits of the adaptive codebook parameter and the adaptive gain codebook parameter, and the second information vector is configured by bits of the linear prediction parameter. Information vector to be obtained, and the adaptive codebook parameter An information vector composed of bits of the adaptive gain codebook parameter, an information vector composed of bits of the noise codebook parameter, and an information vector composed of bits of the noise codebook parameter. The speech decoding apparatus according to claim 14, wherein the information vector is an information vector to be obtained. 16. The method according to claim 13, wherein the first error correction method is a Hamming code method, and the second error correction method is a 1-bit parity code method. A speech decoding device according to claim 1.