JP2855993B2 - Code vector selection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code vector selection system capable of transmitting auxiliary information in voice coding.

[0002]

2. Description of the Related Art Conventionally, the flow of this kind of processing has been shown in FIGS . 7, 8 and 9. FIG . This figure is based on CCITT Recommendation G. 728 (CCITT COM XV-R67, “PART
II OF THE GENEVA MEETING (11-22 NOVEMBER 1991) DRAF
T RECOMENDATION G.728 [CODING OF THE SPEECH AT 16KB
IT / S USING LOW-DELAY CODE EXCITED LINEAR PREDICTIO
N (LD-CELP)] TO BE SUB, MITTED TO RESOLUTION NO.2 AT
In FIG. 7, reference numeral 1 denotes a branching process for judging whether to transmit 1 bit of auxiliary information, and 6 denotes a value of 1 bit of auxiliary information of "0" in FIG. 7 is a process of initializing the code number j of the shape codebook, j is a process of setting j = 0, 8 is a search end value of the code number j of the shape codebook,
processing for setting jend = 64, 9 for initializing the code number j of the shape codebook, processing for setting j = 64, 10 for the search end value of the code number j of the shape codebook, jend = 1
28, 11 is a process for initializing the code number j of the shape codebook, j = 0, and 12 is a process for setting the search end value of the code number j of the shape codebook, jend = 128. 8, reference numeral 15 denotes encoded data output processing, and in FIG. 9, reference numeral 21 denotes one of the LD-CELP methods.
A process of receiving code data, a branching process for determining whether it is time to receive one bit of auxiliary information sent from the encoder side, a process of extracting one bit of auxiliary information sent from the encoder side, Is a process for extracting the shape code book number j from the code data ICHAN, and 31 is a process for extracting the gain code book number i from the code data ICHAN.

First, one bit in the LD-CELP system
The encoding operation when the auxiliary information is transmitted will be described. This will be described with reference to FIG. In processing 1, it is determined whether it is the auxiliary information bit insertion timing. If so, process 6
Judge whether the bit to be inserted is “0” or “1”. If “0”, the start number j and the end number of the shape codebook to be searched in steps 7 and 8 are set to 0 and 64, respectively. If “1”, the start number j of the shape codebook to be searched in steps 9 and 10 is 64, and the end number is 1
28. If it is determined in the processing 1 that the timing is not the auxiliary information bit insertion timing, the processings 11 and 12
The start number j of the shape codebook to be searched is set to 0, and the end number is set to 128. What these processes mean is
When the auxiliary information bit is “0”, the lower half codebook of the shape codebook numbers “0” to “63” is used. When the auxiliary information bit is “1”, the shape codebook number “64” is used.
This means that the upper half codebook of "127" is searched. By these processes, auxiliary information to be transmitted to the IS6 bit of the encoded data ICHAN shown in FIG.
The report bit is inserted in advance.

Next, the process proceeds to encoding process A. This will be described with reference to FIG. The initial value of the root mean square error MIN of the error signal is MI
N=1.0.times.10@15 (a sufficiently large value). The j-th signal vector yj is read from the shape codebook.
Next, the code number i of the gain codebook is initialized, and i =
Processing to set to 0. The i-th gain gi of the code number of the gain code book is read out. The signal vector yj is multiplied by the gain gi, and the vector is set as e. Based on the e vector, a signal vector is synthesized, and this is set as S '. The difference between the input signal vector S and the composite signal vector S 'is taken, and this difference is set as D (D = S-S'). Here, encoding processing B
Proceed to.

[0005] In Fig. 4, D is passed through an auditory weighting filter and is referred to as D '. The root mean square error E of D ′ is calculated. Here, if E is smaller than MIN, MIN
To the value of E, the value of i to IG, and the value of j to IS. If E is equal to or more than MIN, the process proceeds to the next process without doing anything. Add 1 to i and j. If i is not 8 (jump to the processing C), the ith gain gi of the code number in the gain codebook is read again, and i = 8
The above process is repeated until. If i = 8, it is next determined whether j matches the value of jend. If they do not match (jump to processing D), the j-th signal vector yj is read again from the shape codebook, and j = jen
The above processing is repeated until d is reached. If j matches jend, IG is added to ICHAN multiplied by 8 and the value is substituted. At this time, the format of the ICHAN is such that the upper 7 bits are composed of the shape codebook number and the lower 3 bits are composed of the gain codebook number, as shown in FIG. Next, the process proceeds to ICHAN output processing E. In FIG. 8, ICHAN is output as encoded data as it is.

Next, one bit in the LD-CELP system
The decoding operation when the auxiliary information is transmitted will be described with reference to FIG. In a process 21, one encoded data ICHAN sent from the encoder is received. In process 22,
It is determined whether it is time to receive one bit of auxiliary information sent from the encoder side. If the timing is the timing, the IS6 bit of ICHAN shown in FIG. If not, proceed immediately 3
Perform 0. In the process 30, the shape codebook number j is extracted from ICHAN (the INT (...) function is a function for converting fractions below the decimal point to integers).

In step 31, a gain codebook number i is extracted from ICHAN. Thereafter, the process proceeds to a process F which is a decoding process. In FIG. 6, the j-th signal vector yj is read from the shape codebook. Next, the gain gi of the code number i-th in the gain codebook is read out. The signal vector yj is multiplied by the gain gi, and the vector is set as e. Based on this e-vector, a signal vector is synthesized, and this is set as S '. This signal is passed through a post-filter for audibility correction and output as decoded speech.

[0008]

Since the conventional code vector selection method is configured as described above, a certain code
When the search range of the codebook is limited in order to transmit the auxiliary information using the encoding method , the encoding method
If the frequency at which each component in the codebook used in is selected is not statistically uniform and the component within the limited codebook search range was originally a low-frequency component, the decoded speech There was a problem that the quality of the product deteriorated. SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and has as its object to obtain a code vector selection method capable of providing decoded speech of better quality.

[0009]

A code vector selection method according to the present invention investigates the frequency of selection of each component in a codebook used in a certain coding method. When the search range of the codebook is limited in order to transmit the auxiliary information, the search range of the codebook is limited to a range in which frequently selected components are large.

[0010]

In the present invention, even if the search range of the codebook is limited to perform bit stealing and the optimum codebook number is determined, the components in the codebook within this search range are obtained when all codebooks are searched. However, since it is frequently selected as the optimal codebook number, there is a high probability that the optimal codebook number matches the case where all codebooks have been searched, and it is possible to provide better-quality decoded speech.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Voices in 10 languages were input to the LD-CELP codec specified in 728, and the occurrence frequency of the optimal shape code number was investigated. As a result, the ratio of the sum of the optimum shape code number occurrence frequencies between “0” to “63” and the sum of the optimum shape code number occurrence frequencies between “64” to “127” was calculated to be 1: 4. It became the ratio of. Therefore, in this embodiment, one bit
The search range of the shape codebook when transmitting the auxiliary information is fixed between "64" and "127". FIG. 1 is a flow chart of a process on the encoder side of a code vector selection system showing an embodiment of the present invention. Process 1 is the same as the flow chart of the conventional process shown in FIG. 2 is a process for initializing the code number of the shape codebook, j = 64, 3 is a process for setting the search end value of the code number j of the shape codebook, jend = 128, and 4 is a process for initializing the code number of the shape codebook. , J = 0, 5 is the search end value of the code number j in the shape codebook, jend = 1
28.

FIGS. 3 and 4 are flow charts of the LD-CELP process, and the process is the same as that of the conventional example. FIG. 2 is a flowchart of the processing on the decoder side of the code vector selection system according to the embodiment, and the processing 1 is the same as the processing algorithm shown in FIG. 15 is the same as the flow chart of the conventional processing shown in FIG. Reference numeral 13 denotes a process for determining whether the one inserted bit on the transmission side is "0" or "1", reference numeral 14 denotes a process for subtracting 512 from transmission code data, and reference numeral 15 denotes a process for outputting coded data ICHAN. FIG. 5 shows a process on the decoder side of the code vector selection system showing an embodiment of the present invention.
Of a flowchart, the processing 21, 22, 23 is the same as the flowchart of the conventional process shown in FIG. Reference numeral 30 denotes a process for extracting the shape codebook number j from the encoded data ICHAN, and reference numeral 31 denotes a process for extracting the gain codebook number i from the encoded data ICHAN. FIG. 6 shows LD-
It is a flowchart of CELP decoding processing , and processing is the same as the conventional example.

First, the operation on the encoder side will be described. This will be described with reference to FIG. In processing 1, it is determined whether it is the auxiliary information bit insertion timing. If so, the process 2 initializes the shape codebook number j, sets j = 64, and the process 3 ends the search for the shape codebook number, jend = 1.
28. If it is not the above timing, the shape codebook number j is initialized in processing 4 and j = 0, and the search code book number search end value in processing 3 and jend = 1
28. The meaning of these processes is that only the upper half codebook of shape codebook numbers “64” to “127” is searched at the time of inserting the auxiliary information bit, regardless of the value of the bit. It is. Next, the process proceeds to the encoding process A. The encoding process A is the same as the conventional example.

[0014] Next Oite supplementary information bit inserting process is described in FIG. Process 1 smell again determines whether auxiliary information bit insertion timing Te. If so, it is determined in step 13 whether the auxiliary information bit is "0",
If "0", subtract 512 from ICHAN and re-ICH
Substitute into AN and perform processing 15 . On the other hand, if "1", the process 15 is performed as it is. In step 15 , ICHAN is output as encoded data. By the processing of FIG.
The auxiliary information bit is inserted into the IS6 bit of the transmission data ICHAN.

Next, the operation of the decoder will be described with reference to FIG. In a process 21, one encoded data ICHAN sent from the encoder is received. In a process 22, it is determined whether it is a timing to receive the auxiliary information bits sent from the encoder side. If it is the timing, processing 23
Then, the IS6 bit of ICHAN of FIG. In step 24, the shape codebook number j is extracted from ICHAN. In step 25, a gain codebook number i is extracted from ICHAN. In process 26, it is determined whether the shape codebook number j is smaller than 64,
If it is smaller, 64 is added to j in step 27, and the process proceeds to decoding process F. If it is not smaller, the process directly proceeds to decoding process F. In the process 26, the search range of the shape codebook on the encoder side is "6".
Since it is limited to 4 "to" 127 ", this is a process of adding an offset of 64. On the other hand, if it is determined in the process 22 that it is not the timing to receive the auxiliary information bit, the processes 28 and 29 are performed and decoding is performed. The process proceeds to the process F. Here, the process 28 corresponds to the process 24, and the process 29 corresponds to the process 25. The decoding process F is the same as the conventional example.

In the above embodiment, the LD-CELP codec has been described as an example. However, the same effects as those of the above embodiment can be obtained by using a speech coding system that performs a coding process using a codebook. Further, in the above embodiment, the codebook is halved to secure one bit of auxiliary information transmission bit . However, if the codebook is further reduced to 1 / 2n, n bits of the auxiliary information are transmitted .
An auxiliary information transmission bit can be secured, and the same effect as in the above embodiment can be obtained.

[0017]

According to the code vector selection method of the present invention, in a certain coding method, the frequency at which each component in the code book is selected is investigated, and the search range of the code book is limited in order to transmit auxiliary information. When searching, the search range of the codebook is limited to the range where the components that are frequently selected are large, so even if the optimal codebook number is determined, the components in the codebook of this search range are all codes. Even when a book is searched, it is frequently selected as the optimal codebook number, so there is a high probability that the optimal codebook number will match the search for all codebooks.
Better quality decoded speech can be provided.

[Brief description of the drawings]

FIG. 1 is a flowchart of a shape codebook search range determination process in a code vector selection method for transmitting auxiliary information in an LD-CELP codec according to the present invention.

FIG. 2 shows a code vector selection scheme for transmitting auxiliary information in an LD-CELP codec according to the present invention;
It is a flowchart of transmission auxiliary information bit insertion processing.

FIG. 3 shows a processing flow of an encoder according to the LD-CELP method.
FIG .

FIG. 4 is a flow of processing performed by an encoder using the LD-CELP method.
FIG .

FIG. 5 is a flowchart illustrating a reception shape codebook search range determination process and a reception complement in a code vector selection method for transmitting auxiliary information in an LD-CELP codec according to the present invention;
It is a flowchart which shows the Example of auxiliary information bit extraction processing.

FIG. 6 shows the flow of processing performed by a decoder based on the LD-CELP method.
FIG .

7 is a flow <br/> view of a shape codebook search range determination process and transmit auxiliary information bit insertion processing in the code vector selection method in LD-CELP codec according to the conventional example.

FIG. 8 is a flowchart of a code data output process according to a conventional example.

9 is a flow diagram illustrating an embodiment of a receiver auxiliary information bit extraction process in a code vector selection method in LD-CELP codec according to the conventional example.

FIG. 10 is a diagram illustrating a bit configuration of encoded data ICHAN.

[Explanation of symbols]

1 Branch processing for determining whether it is time to transmit one bit of auxiliary information 2 Initializing the code number of the shape codebook, j = 64
3 Search end value of code number j of shape codebook, je
Processing to set nd = 128 4 Initialization of code number of shape codebook, processing to set j = 0 5 Search end value of code number j of shape codebook, je
Processing for setting nd = 128 6 Branch processing for determining whether the value of one bit of auxiliary information is “0” or “1” 7 Initializing code number j of shape codebook, j = 0
8 Search end value of code number j of shape codebook, je
Processing to set nd = 64 9 Initialize code number j of shape codebook, j = 6
Processing to be 4 10 Search end value of code number j of shape codebook, j
End = 128 processing 11 Initialize code number j of shape codebook, j =
Processing to be set to 0 12 Search end value of code number j of shape codebook, j
End = 128 processing 13 1 bit inserted on the transmitting side is “0”
Processing to determine whether it is "1" 14 Processing to subtract 512 from transmission code data 15 Processing to output encoded data 21 Processing to receive 1 code data of LD-CELP system 22 Receive 1 bit of auxiliary information sent from the encoder side Branch processing to determine timing 23 Processing to extract 1 bit of auxiliary information sent from encoder 24 Processing to extract shape codebook number j from ICHAN 25 Processing to extract gain codebook number i from ICHAN 26 Shape It is determined whether or not the codebook number j is smaller than 64. If the codebook number j is smaller, the process proceeds to the process 27. If the codebook number j is not smaller, the process proceeds to the decoding process F as it is. Correspondence 30 Shape code book number j from code data ICHAN
31. Processing to extract the gain code book number i from the code data ICHAN

Continuation of the front page (56) References JP-A-61-258300 (JP, A) JP-A-3-222522 (JP, A) JP-A-4-342300 (JP, A) JP-A 2-145078 (JP) JP-A-3-113571 (JP, A) JP-A-3-243996 (JP, A) JP-A-5-191293 (JP, A)

Claims (1)

(57) [Claims] 1. A code base vector selection method in a speech coding apparatus using a codebook, the codebook used in the above speech encoding apparatus
At least a group of frequently selected code numbers
A code book that can delete one or more bits of the code number
Procedure to newly provide and insert auxiliary information for transmission
A search step for limiting the range of the new code book only, and a step of inserting and transmitting auxiliary information for bits obtained by removing the number of bits of the code number. Vector selection method.