JPH01161930A - Correction method for adaptive bit assignment - Google Patents

Abstract

PURPOSE:To attain high speed correction without increasing the hardware by applying correction to total bit number (bk) one by one bit each when the total bit number (bk) requiring the correction does not exceed a predetermined value and applying the correction of plural bits to the total bit number bk when the predetermined value is exceeded. CONSTITUTION:When the total bit number (bk) requiring the correction does not exceed a predetermined value, the correction of the total bk is applied one by one bit each and when the total bit number (bk) requiring the correction exceeds a predetermined value, the correction of the plural bits is applied to the bit bk. When the total bit number requiring the correction exceeds a predetermined value in this way, since the correction for plural bits is applied to the total bit bk in a batch at each band, the correction is attained at a high speed without incurring the increase in the hardware even with a large correction quantity.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a correction method for adaptive bit v1 in information compression.

(Prior art) Generally, when performing digital transmission, recording, etc., analog information such as images and audio is digitized, information is compressed to eliminate redundancy, and coded (
ωft).

FIG. 3 illustrates this m-childization. The horizontal axis shows the input value, the vertical axis shows the output value, and the number attached to each stage shows the Q childization code. This example shows the case of encoding into 4 bits, and there are 24 codes. If the intervention force value is 0.34, the code is 3, the restored output value is 0.32, and the Q child conversion error ax in this case is 0.0.
It becomes 2. When quantization and inverse quantization are performed in this way, Q-childization J u generally occurs.

Conventionally, in order to minimize this O-column conversion error, a method has been considered in which the number of m-column bits is varied according to the power of the input signal. The case of voice Shintan will be exemplified below.

The distortion D in the case where the input signal BX(n), the output signal BX(n), and the input signal BX(n) are processed as frames every N points is given as follows.

Next, speech generation uses phonological information representing the structure, and the orthogonal transform of the signal from which redundancy has been removed is divided into frequency bands, and Q
Input it to the child generator and perform Φft. Here, the number of data input to the r child generator per time is M, the number of frequency division bands is L, and the number of bits allocated to each band is bk (k=o,...
・, [-1) Doshi, B → Average information gain per nonpull,
If gk is the power of each band based on phonetic information, distortion D
/j-R can be made small by performing adaptive bit allocation as shown in (2).

The value calculated by equation (2) is not necessarily a positive integer, and must be rounded to an integer. In addition, in reality, for the purpose of preventing extreme bias in the number of R11 pins,
Corrections were made to bk in order to limit the minimum and maximum values of k and make the total number of bits of bk a constant value.

Next, such adaptive bit 'fJ' and its correction method will be explained according to flowcharts 1 to 1 shown in FIGS. 4 to 6. The pair 4 diagram is a routine that calculates the number of bits BK to be allocated to each band, and calculates the total number of bits F, -1g1.

In steps 401 and 402, the power of each band is determined based on the phonetic information. Next, the input power (starting from Jk) is calculated (step 403). Next, in step 404, the processes of steps 405, 406, and 407 are repeated L times. In step 405, the number of bits bk to be allocated to each band is calculated. In step 406, the spoken bk is limited to fall within a preset lower limit B1 and upper limit 1fiB2.In step 407, the limited bk is rounded off to an integer.

The processing from steps 408 to 410 is to reduce the total number of bits. In step 408, an initial value O is set to a variable SUMlb representing the total number of bits. In step 409, the process in step 410 is repeated L[i] times. and step 409, step 410
The total number of bits SLIMIb is calculated by the process.

In step 411, the total number of bits suM
It is determined whether or not tb matches the preset total bit number 1"BIT, and if they match, the process ends. If they do not match, the total bit number SUMIb set in 81 times is set. It is determined whether or not the total number of pits T B I T exceeds (step 412 >, SUMIb
>If TBIT, 4 knob routine COR8LIB (
The processing is performed by step 413), and SUMIb
If <TB IT, processing is performed by subroutine C0RADD (step 414).

FIG. 5 is a flowchart showing the contents of blue routine COR8UB. This subroutine selects the total number of sifted 1 to number SUMIb set (・number T B I
Since it is larger than T, a correction is made to reduce the number of bits applied to the predetermined band:9.

That is, in step 501, the correction factor TDIFF is totaled to 0 by TDIF=SLIMIb-TBIT. Then, the processing from step 501a to step 510 is performed using this correction rl=1' D I F
Repeat until F becomes zero. In step 501a, an appropriate maximum positive value is assigned to the variable MINIB. Steps 502 to 507 are processes for finding the band with the minimum (bk - 1bk).
The processing up to 07 is repeated L times. It is determined whether the integerized pitch 1-number lbk matches the lower limit (di F31 (su-)-tub 503), and Ten = bk-1bk is added to the platform that does not match (step 504). ). Next, the variable Tc1+p
The smaller value of the variable MINlb and the variable MINlb is assigned to MINlb (step 505). Variable MINlb is variable 1en
If J matches +p (step 506), set k to variable MINk.
is substituted (step 507).

This MINk represents the number of the minimum band (bk-1bk).

When the band with the minimum (bk-1bk) is found, the pitch I/number lbMINk applied to the minimum band is 1
The bit is subtracted (step 508) and then the correction fM-
1 is subtracted from rDIFF (step 509). Then, in step 510, the correction ω]IFFtfiU
It is determined whether or not it matches the mouth, and the processes from step 501a to step 509 are repeated until it matches zero. .

FIG. 6 is a flowchart showing the contents of Sunbruven C0RADD. This subroutine is t) The total number of bits issued SLIMIb is the total number of bits set TBIT
Therefore, a correction is made to increase the number of bits allocated to a predetermined band.

That is, in step 601, the correction mTDIFF is calculated by TOIFF=TBIT-8UMIb. , and repeats the processing from step 601a to step 610 until this corrected mTD I FF becomes zero. In step 601a, an appropriate negative minimum value is assigned to the variable MAXIb. Steps 602 to 607 are processes for finding the band with the minimum (t)k -1bk ). That is, step 602 causes step 603 to step 607
The processes up to this point are repeated L times. Integerized pi-
It is determined whether the number lbk matches the upper limit B2 (
Step 603), and if they do not match, TOIIEl=bk-1bk is calculated (step 604). Next, the larger value of the variable Tenp and the variable MAXIb is assigned to MAXIb (step 605). If MAXIb matches the variable Te1jp (step 606), k is assigned to the variable MAXk (step 607). This MAXkG,t
(bk - 1bk) represents the number of the 8 largest bands.

When the maximum band (bk-1bk) is found, 1 bit is added to the number of bits 1bmaxk allocated to the maximum band (step 608), and then the corrected flTDIF
1 is subtracted from F (step 609). and 610
In step 601a, it is determined whether the corrected ωTDIFF matches zero or not, and the processes from step 601a to step 609 are repeated until it matches zero.

(Problems to be solved by the invention) In this way, the present invention of the conventional adaptive bit allocation
-1bk) is the minimum (maximum), then 1 bit is subtracted from the number of bits 1b allocated to this band (
If the amount of correction is large and only one bit can be corrected at one time, the processing time will be longer and a certain amount of processing will be required to process within one period. There was a problem with the need for hardware.

The present invention has been made in view of these problems, and its purpose is to provide a correction method for adaptive bit allocation that allows correction to be made at high speed without requiring additional hardware, even when the correction Q is large. There is a particular thing.

[Structure of the invention]

(First stage for solving the problem) In order to achieve the above object, the present invention combines multiple pieces of data into one
When dividing W'J into frequency bands, the number of bits blt (k = o, ..., L-1) to be allocated to each band is calculated using a predetermined algorithm and is determined in advance as the total number of bits bk. In the correction method of correcting bk using a predetermined method if the total number of bits that require correction does not match the predetermined total number of bits, if the total number of bits that require correction does not exceed a predetermined value, bk is corrected one bit at a time. If the total number of bits required for bk exceeds a predetermined value, a plurality of bits are corrected for bk.

(Function) In the present invention, when the total number of bits that require correction exceeds a predetermined value, multiple bits can be corrected at once for bk in each band, so correction can be performed at high speed. Also, there is no need to worry about the increase in hardware.

(Example) An example of the present invention will be described in detail below based on the drawings. In the adaptive bit allocation correction method according to an embodiment of the present invention, the main routine is the same as the one not shown in FIG.
Since the contents of the DD are different from the conventional ones, these subroutines will be explained.

FIG. 1 is a flowchart showing the contents of subroutine COR8OB. The total number of pits TDIFF that requires correction in step 101 is TD I FF=SLJB l
b - Calculated by TB IT. In step 102, if J> and TOIFF is larger than the number of frequency division bands, multiple bits are corrected at once. J, that is, the variable MINlb, is set to a positive maximum value, and zero is given as the initial value of the correctable number n (step 10
3).

The processing from step 104 to step 107 is a process of finding the minimum value of Ibk when Ibk does not match the lower limit 181, and extracting the correctable number n. That is, in step 104, the processes from steps 105 to 107 are repeated L times. i in step 105
It is determined whether or not τlbk matches the lower limit value B1, and if they do not match, the smaller value of Ibk and MINlb is assigned to MINlb (step 106);
Then, 1 is added to the correctable number “1” (step 1
07).

Next, in step 108, the number of bits IX that can be corrected at one time per band is determined by IX=MIN lb - Bl, and in step 109, the total number of pits that can be corrected at one time TIX is determined as TIX=IX-n. If TIX>TDIFF, IX=TDIFF/n is calculated in step 116, a predetermined truncation is performed in step 117, and then the process returns to step 109.

TIX≦TDIFF '3 c. From step 111 to step 113, the number of bits IX that can be corrected at one time is subtracted from the number of bits lbk allocated in each band. Zunawa also performs step 111 and step 112.
The process of 113 is repeated L times. If Ibk does not match the lower limit B1 (step 112Mbk to I
X is drawn. (Step 113). In this way, in each band, the number 1x of bits that can be corrected by one or more bits is subtracted from Ibk.

Next, in step 114, the total number of pits T that needs to be corrected
The total number of pits TIX that can be corrected at once is subtracted from DIFF, and in step 115 it is determined whether TDIFF is zero. If it is not zero, the process returns to step 102 and the above steps are not repeated.

If TDIFF is smaller than L in step 102, the routine moves to the routine shown in FIG. 1(C). This routine corrects one bit at a time and is similar to the process shown in FIG. 5, so a redundant explanation will be omitted.

FIG. 2 is a flowchart showing the contents of the service routine COR800. The total number of pits TDIFF that requires correction in step 201 is TDIFF=TBIT-8L
4 by IMIb. TDIF in J3 at step 202
If F is larger than the number of frequency subbands, -
3, where multiple bits are corrected at a time, that is, the variable MA
A negative minimum value is set for XIb, and zero is set as the initial value of the number n that can be corrected! ] is obtained (step 203).

In the process from step 204 to step 207, if Ibk does not match the upper limit B2, the maximum value of elbk is determined and the correctable number n is determined. That is, by step 204, steps 205 to 207
The processes up to this point are repeated L times. Step 205
At step d3, it is determined whether or not the Ibk upper limit value B2 matches, and if they do not match, the larger value of Ibk and MAXIb is substituted into MAXIb (step 206).

Then, 1 is added to the correctable number n (step 20
7).

Next, in step 208, the number of bits that can be corrected at one time per band IX is determined as IX=82-MAXIb, and in step 209, the number of bits that can be corrected at one time TIX is determined as TIX=IX-rl. If TIX>TOIFF, the process goes to step 216 to calculate IX=TD IFF/n, and after executing a predetermined truncation process in step 217, the process returns to step 209.

]Ix≦■D■FF, in steps 211 to 213, the number of bits EX that can be corrected at one time is added to the number of bits lb allocated in each band. That is, by step 211, steps 212 and 213
The process is repeated L times. Ibk is the upper limit B2
If they do not match (step 212), IX is added to Ibk (step 213). In this way, the number of bits that can be corrected by one or more pits is added to Ibk in each band. Next, in step 214, the total number of bits that can be corrected at once is subtracted from the total number of bits TDIFF that requires correction, and in step 215, TD
It is determined whether FF is zero. If it is not zero, the process returns to step 202 and the above operations are repeated.

If TDIFF is smaller than L in step 202, the routine moves to the routine shown in FIG. 2(C). This routine corrects one pit at a time and is similar to the process shown in FIG. 6, so a redundant explanation will be avoided.

Thus, in this embodiment, in the subroutines COR5LIB and C0RADD, the total number of bits TOI that needs to be corrected
When FF is larger than the number of frequency division bands, for the bits vllbk allocated in each band,
Since the correction is performed by 1 pip or more, it is possible to perform the correction process quickly.

〔Effect of the invention〕

As described above in detail, according to the present invention, even if the correction Q is large, the correction can be performed at high speed without requiring additional hardware.

[Brief explanation of the drawing]

Fig. 1 is a flowchart showing the contents of subroutine COR8UB, Fig. 2 is a flowchart showing the contents of 4J bluebun C0RADD, Fig. 3 is an explanatory diagram showing ωft,
Figure 4 is a 70-chart showing the contents of the main loop,
FIGS. 5 and 6 do not show the contents of the conventional subroutines COR8UB and C0RADD, but are flowcharts ↑I-. Figure 1 (b) Mair Routine Figure 4 (C1) Figure 4 (C) Figure 4 (d)

Claims (1)

[Claims] When dividing a plurality of data into L frequency bands and vector quantizing each band, the number of bits bk (k=0,...L-1) to be allocated to each band is calculated using a predetermined calculation method. In the case where the total number of bits bk calculated by and the predetermined total number of bits do not match, bk is corrected by a predetermined method. In the correction method, the total number of bits that need to be corrected does not exceed a predetermined value. A correction method for adaptive bit allocation, characterized in that if the total number of bits that require correction exceeds a predetermined value, then correction is performed on bk one bit at a time.