CN101308657B - Code stream synthesizing method based on advanced audio coder - Google Patents

Abstract

The invention discloses a code stream synthesis method based on an advanced audio encoder, comprising the following steps: A, quantization encoding: perform quantization encoding processing on the code stream to be encoded to obtain a plurality of encoding subbands; B, codebook merging: calculation The reference bit consumption value when different codebooks are used for each coding subband, and determine the codebook used by each coding subband: according to the coding sequence, for the first coding subband, select the codebook with the smallest reference bit consumption value in each codebook As the use of codebooks; for each coded subband after the first coded subband, if the quantization spectrum of the current coded subband is not all 0, select each codebook so that the total of the current coded subband and the previous coded subband Refer to the codebook with the smallest bit consumption value as the codebook for use; C. Code stream synthesis: For each coded subband, select the codebook for use determined in step B for encoding, and complete the code stream synthesis process. The invention reduces the bit overhead of entropy coding and improves the coding efficiency of the advanced audio coder.

Description

A Code Stream Synthesis Method Based on Advanced Audio Encoder

technical field

The present invention relates to the advanced audio coding (Advanced Audio Coding, AAC) of MPEG-4 (a kind of dynamic picture standard formulated by the Motion Picture Experts Group), in particular to a code stream synthesis method based on an advanced audio coder.

Background technique

Audio coding technology is at the heart of multimedia applications such as digital audio broadcasting, music dissemination over the Internet, and audio communications, which benefit greatly from improvements in the compression performance of audio encoders. As a kind of lossy transform domain coding, perceptual audio coder is the modern mainstream audio coder. The perceptual audio coder relies on the study of the auditory perception characteristics of the human ear, performs perceptual shaping on the coding noise, and obtains a coding effect with a lower perceptual quality degradation with a higher coding compression ratio. In the perceptual audio coding algorithm, after the transform domain spectrum of the audio signal is quantized, an entropy encoder is needed to encode the quantized spectrum to reduce information redundancy. As a perceptual audio coder, in MPEG-4AAC, quantization is in units of subbands, and the MDCT (modified discrete cosine transform) spectrum in the subbands is quantized to obtain integer quantization spectra, and the integer quantization spectrum is in units of subbands Entropy coding using Huffman codebook. The Huffman codebook in MPEG-4AAC includes the codeword codebook HW _k and the code length codebook HL _k , k represents the serial number of the codebook, ranging from 1 to 11, representing 11 different Huffman codebooks, which are listed in in FIG. 1. The Huffman codeword is obtained by retrieving the codeword codebook, and the length of the Huffman codeword is obtained by retrieving the code length codebook, that is, the code length.

Table 1 Huffman codebook and its properties in MPEG-4AAC

codebook serial number symbol dimension Maximum absolute value The total number of 1 have 4 1 81 2 have 4 1 81 3 none 4 2 81 4 none 4 2 81 5 have 2 4 81 6 have 2 4 81 7 none 2 7 64 8 none 2 7 64 9 none 2 12 169

10 none 2 12 169 11 none 2 15 289

In Table 1, a sign means that the codebook directly encodes the integer quantized spectrum; an unsigned means that the codebook only encodes the absolute value of the integer quantized spectrum. The n-dimensional codebook represents that n integer quantization spectra correspond to one Huffman codeword; the maximum absolute value represents the maximum absolute value of the integer quantization spectrum in the subbands that can be coded using the codebook. To retrieve the code length of the entropy code, first obtain the maximum absolute value t _max of the integer quantization spectrum in the sub-band, and sequentially form the four-dimensional vector [t ₁ t ₂ t ₃ t ₄ ] of four adjacent integer quantization spectra in the sub-band Use it as an index to search and accumulate to obtain the total number of code lengths required for entropy encoding of different codebooks. Among them, the index of the signed codebook should be added with the maximum absolute value as the offset, while the index of the unsigned codebook should be added with 1. The four-dimensional code book uses [t ₁ t ₂ t ₃ t ₄ ] as the index search for each dimension, and the two-dimensional code book uses [t ₁ t ₂ ] and [t ₃ t ₄ ] to search twice. The traditional codebook retrieval process is judged according to t _max as follows:

t _max ＝1, search 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 number books

t _max = 2, search 3, 4, 5, 6, 7, 8, 9, 10, 11 number books

2<t _max □4, search 5, 6, 7, 8, 9, 10, 11 number books

4<t _max □7, search 7, 8, 9, 10, 11 number books

7<t _max □12, search 9, 10, 11 number book

12<t _max □15, retrieve book number 11

16 ￡ t _max , search for the number 11 book, start the overflow protection mechanism in the overflow protection mechanism, first use 16 to replace the integer spectrum t greater than 16 as the index to search when t _max ³ 16, and then follow the Huffman code word , sequentially add escape sequences for indexes greater than or equal to 16. An escape sequence consists of the following three parts:

[escape_prefix][escape_separator][escape_word]

Where [escape_prefix] is N '1', [escape_separator] is a '0', [escape_word] is an unsigned integer of N+4 bits, t satisfies t=2 ^N+4 +[escape_word], increase The number of bits is 2int[log2(t+1)+1]-5.

Obtain the bit consumption reference value of the nth subband integer quantization spectrum using the m number book by searching the code length, and record the bit consumption reference value in the reference bit consumption matrix refBits with the codebook number and the subband number of the integer quantization spectrum as the row and column parameters (m, n). refBits is used as the basis for determining the sequence number of the codebook finally used by each subband. The code stream synthesis of AAC requires additional bits to store the codebook number used by each subband. Starting from the first subband, it uses 4 bits to represent the codebook number, and 5 bits to represent the subbands that continuously use the codebook. Band number, if the number of subbands using the serial number book reaches or exceeds 31, then add 5 bits to represent the value of the number of continuous subbands minus 31. In this way, usually, only 9 bits are added to represent the overhead of the codebook serial number. In fact, because the maximum value of the sub-band quantization spectrum is not the same, and different codebooks are used for the same sub-band, the number of bits consumed by the quantization spectrum entropy coding is very different, and the use of the same codebook for the sub-bands cannot save energy when the code stream is synthesized. It is a complex optimization process to determine the optimal sub-band sequence number of each sub-band, and to combine consecutive sub-bands using the same codebook for bit overhead. Traditional methods include, regardless of optimization, selecting the codebook with the least bit consumption for each subband; merging fixed-length codebooks, forcing three or more adjacent subbands to use the same codebook with the least bit consumption, etc. These methods still consume a large number of bits, which affects the efficiency of entropy coding.

Contents of the invention

The technical problem to be solved by the present invention is to provide a code stream synthesis method based on an advanced audio encoder, which can better optimize the codebook merging process, thereby improving the code stream synthesis effect and further improving the performance of the entire audio encoder.

In order to solve the above technical problems, the present invention adopts the following technical solutions:

A code stream synthesis method based on an advanced audio encoder, comprising the steps of:

A. Quantization coding: perform quantization coding processing on the code stream to be coded to obtain multiple coding subbands;

B. Codebook merging: calculate the reference bit consumption value when each coding subband uses different codebooks, and determine the codebook used by each coding subband: according to the coding order, for the first coding subband, select the reference in each codebook The codebook with the smallest bit consumption value is used as the codebook; for each coded subband after the first coded subband, if the quantization spectrum of the current coded subband is not all 0, select each codebook so that the current coded subband and The codebook with the smallest total reference bit consumption value of the last coded subband is used as the codebook;

C. Code stream synthesis: for each coded sub-band, select the codebook determined in step B for encoding, and complete the code stream synthesis process.

Each coding subband selection codebook after the first coding subband in the step B is determined by the following principles:

Obtain the reference bit consumption value of each codebook used in the current coding subband. If the codebook with the smallest reference bit consumption value is the same as the codebook used in the previous coding subband, use this codebook for the current coding subband. Otherwise, proceed as follows: Calculate the total reference bit consumption value of the current coded subband and the previous coded subband in two ways, and choose to use the codebook in a way that minimizes the total reference bit consumption value of the current coded subband and the previous coded subband:

1) Take the selected codebook of the previous coding subband as the candidate codebook of the current coding subband, and the total reference bit consumption value is equal to the sum of the reference bit consumption values of the two coding subbands;

2) In all other codebooks except the selected codebook for the previous coding subband, select the codebook that minimizes the sum of the reference bit consumption values of the two coding subbands as the current coding subband and the previous coding subband The total reference bit consumption value is equal to the sum of the reference bit consumption values of the two coded sub-bands. If the previous coded sub-band and the previous coded sub-band use the same codebook, the total reference bit consumption value It is equal to the sum of the reference bit consumption values of the two coded subbands plus the codebook switching overhead.

3) Select the codebook with the smallest reference bit consumption value of the current coding subband as the candidate codebook, and the total reference bit consumption value is equal to the sum of the reference bit consumption values of the previous coding subband and the current coding subband plus the codebook switching overhead.

In the codebook merging process of the step B, if the quantized spectrum of one coded subband or multiple consecutive coded subbands is all 0, count the coded subband or the multiple consecutive coded subbands using the previous one that is not all 0 The reference bit consumption value or the sum of the reference bit consumption values encoded by the codebook used by the coding subband, if the reference bit consumption value or the sum of the reference bit consumption value is greater than the codebook switching overhead, the coding subband or multiple consecutive Coded sub-bands are coded using the 0 codebook; otherwise, codebook codes used by the previous coded sub-band that is not all 0s are used.

The reference bit consumption value is pre-stored in the two-dimensional array refBits, and refBits(m, n) is the element value of the mth row and the nth column in the array refBits, which indicates that the integer quantization spectrum of the nth subband uses m number book reference Bit consumption value, the initial value of refBits(m, n) is set to a maximum of 65535.

If the integer quantization spectrum of the current coded subband is all 0, the codebook reference bit consumption value corresponding to the subband is refBits(m, n)=bw _sfb (n)*CW ₀ (m), where bw _sfb and CW ₀ is a one-dimensional array, bw _sfb (n) is the value of the nth element in the array bw _sfb , which represents the bandwidth of the coding subband n, and CW ₀ (m) is the value of the mth element in the array CW ₀ , which is represented by The number of bits per spectral line consumed by codebook m to encode an integer quantized spectrum of 0 values.

Also includes array preprocessing, the array preprocessing is carried out using the following formula: if refBits (m, n+1) = 65535, then refBits (m, n) = refBits (m, n) + C, where refBits (m, n) is the reference bit consumption value of the current coding subband, refBits(m, n+1) is the reference bit consumption value of the next coding subband, and C is the codebook switching overhead.

The one-dimensional array CW ₀ =[0.250.750.2510.520.52.50.532].

The codebook is a Huffman codebook, which contains 11 serial numbers.

The reference bit consumption value is obtained by searching the Huffman codebook with the integer quantization spectrum of the coded subband as an index.

In the present invention, by ensuring the minimum total reference bit consumption value during encoding, for each coded subband, the codebook used when the total reference bit consumption value of the previous coded subband is the smallest is selected to determine the codebook used by each coded subband. Codebook, using this first-order backtracking method to optimize the codebook merging process in codestream synthesis, thereby reducing bit overhead, improving the effect of codestream synthesis, and thus improving the coding efficiency of advanced audio encoders.

Further, when the integer quantization spectrum of the coded subband is all 0, compare the different total reference bit consumption values in the two cases where the codebook used by the previous coded subband and the 0 codebook are used for the coded subband, and choose to use A codebook with a smaller total reference bit consumption value is used as a codebook for encoding subbands whose integer quantization spectrum is all 0.

Furthermore, the reference bit consumption value array is preprocessed, and the codebook switching overhead is added to the backward isolated array unit (that is, the rear column is 65535). By considering the first-order backward situation, the codebook merging is improved. optimization effect.

Description of drawings

Fig. 1 is the processing flowchart of the specific embodiment of the present invention;

Fig. 2 is under 100 frames of 80kbps stereo encoding, various codebook merging methods in the prior art and the codebook merging method adopted in the specific embodiment of the present invention are compared to the bit consumption comparison effect of the minimum value bit consumption;

Fig. 3 is under 100 frames of 128kbps stereo coding, various codebook merging methods in the prior art and the codebook merging method adopted in the specific embodiment of the present invention are compared with respect to the bit consumption effect of the minimum value bit consumption;

Fig. 4 is under 100 frames of 192kbps stereo encoding, various codebook merging methods in the prior art and the codebook merging method adopted in the specific embodiment of the present invention are compared with respect to the bit consumption effect of the minimum bit consumption;

Fig. 5 is under 100 frames of 256kbps stereo encoding, various codebook merging methods in the prior art and the codebook merging method adopted in the specific embodiment of the present invention are compared with respect to the bit consumption effect of the minimum value bit consumption;

Fig. 6 is a comparison effect of various codebook merging methods in the prior art and the codebook merging method adopted in the specific embodiment of the present invention with respect to the minimum bit consumption under different bit and multiple types of audio coding.

Detailed ways

The present invention will be described in detail below in conjunction with specific embodiments with reference to the accompanying drawings.

As shown in Figure 1, in the code stream synthesis of MPEG-4AAC, it is necessary to select different codebooks for encoding the multiple coding sub-bands formed after the code stream is quantized. This is called the codebook merging process. Whether codebook merging is optimized directly affects the effect of code stream synthesis, thereby affecting the coding efficiency of the entire encoder. The processing procedure of the specific embodiment of this method comprises:

1. Quantization coding: perform quantization coding processing on the code stream to be coded to obtain multiple coding subbands;

2. Codebook merger:

1. Preset the two-dimensional array refBits(m, n), where n represents the nth coding subband of the integer quantization spectrum obtained after flow coding of the code to be coded, and m represents the serial number of the codebook used by each coding subband during coding ;refBits(m,n) represents the reference bit consumption value of the nth sub-band integer quantization spectrum using the m number book; the initial value of each value in the array is set to the maximum value of 65535. In this embodiment, a Huffman codebook is used, including a codebook with 11 serial numbers.

2. Calculate the reference bit consumption values of all coded subbands in each codebook and store them in the two-dimensional array refBits(m, n).

3. When a subband does not need to be coded (that is, the quantized spectrum of the subband is all 0), the reference bit consumption value of the codebook corresponding to the subband is refBits(m, n)=bw _sfb (n)*CW ₀ ( m), where bw _sfb (n) is the bandwidth of the currently coded subband n, and CW ₀ (m) is the number of bits per spectral line consumed by encoding an integer quantized spectrum of 0 values with the codebook m, which is called the null subband With jump avoidance principle.

4. Perform preprocessing on the array, considering that codebook switching needs to increase the overhead of C bits. The preprocessing method is: if refBits(m, n+1)=65535, then refBits(m, n)=refBits(m, n)+C. That is, if the following subband does not use the codebook of the current subband, the reference bit consumption value of the codebook corresponding to the current subband is added by C. The codebook switching overhead C uses 4 bits to represent the codebook sequence number, and 5 bits to represent the number of subbands that continuously use the sequence numberbook.

5. Through the first-order backtracking codebook merging method, select the codebook combination with the smallest bit overhead.

5.1 According to the coding sequence, for the first coding subband, select the codebook with the smallest reference bit consumption value from the 11 codebooks in the Huffman codebook.

5.2 For each coded subband after the first coded subband, the componentized spectrum is all 0 or not all 0;

5.3 The case where the quantization spectrum of the current subband k is not all 0:

5.3.1 If the codebook with the smallest reference bit consumption value of the current subband in each sequence number book is the same as the codebook selected for the previous subband, then select the sequence number book as the codebook used by the current subband.

5.3.2 Otherwise, calculate the total reference bit consumption value of the current subband and the previous subband in the following three candidate modes, and choose the mode with the smallest total reference bit consumption value:

5.3.2.1 Take the codebook that has been selected for the previous subband as the candidate codebook for the current subband (two subbands use the same codebook), and the total reference bit consumption value is equal to the sum of the reference bit consumption values of the two subbands.

5.3.2.2 Take the same candidate codebook for the current subband and the previous subband, traverse all codebooks with serial numbers except the codebook selected for the previous subband, and select the one with the smallest sum of their two reference bit consumption values The codebook is used as a common candidate codebook for two subbands (two subbands use the same codebook). In this case, the candidate codebook used by the previous sub-band has changed, so if the previous sub-band and the previous sub-band use the same codebook, the total reference bit consumption value should be added to the codebook switch The cost of C.

5.3.2.3 Select the codebook that minimizes the reference bit consumption value of the current subband as the candidate codebook in each serial number book (two subbands use different codebooks), and the total reference bit consumption value is equal to the reference bit consumption value of the previous subband value plus the minimum reference bit consumption value of the current subband, plus the codebook switching overhead C.

5.4 The case where the quantized spectrum of the current subband k is all 0:

5.4.1 For such a situation, in the prior art, the number 0 book is generally used directly, but such selection causes codebook switching, thus increasing the codebook switching overhead. However, in this embodiment, the codebook of the previous subband (quantization spectrum is not all 0) is used to avoid codebook switching. Use the codebook of the previous subband to quantize the number of bits consumed by the 0-value integer spectrum as the reference bit consumption value of the current subband, zeroBits=refBits[BookIndex(k-1), k], BookIndex(k-1) means The codebook of the previous subband.

5.4.2 If the quantized spectrum of the subsequent subband k+1 is still 0, the codebook of the previous subband whose quantized spectrum is not all 0 is still used for quantization, and the number of bits consumed by the codebook to quantize the integer spectrum with a value of 0 is used as The reference bit consumption value of the current subband. At the same time, the reference bit consumption value is accumulated, that is, the sum of the reference bit consumption values of multiple consecutive coded subbands whose quantization spectrum is 0 is calculated. zeroBits=zeroBits+refBits[BookIndex(k-1), k+1]. zeroBits is the total reference bit consumption value for encoding subsequent consecutive subbands whose quantization spectra are all 0 with the codebook of k-1 subbands.

5.4.3 Repeat step 5.4.2 until the quantized spectrum of subbands is not all 0.

5.4.4 If zeroBits>C, code the subband whose quantization spectrum is all 0 with the 0 number book, and C is the overhead bit for codebook switching.

5.4.5 If zeroBits<=C, use the codebook of k-1 subbands to encode the subsequent continuous subbands whose quantized spectra are all 0.

5.4.6 If the quantization spectrum of subband k+1 is not all 0, that is, only the quantization spectrum of one coded subband of the current subband k is all 0, then if the reference bit consumption value of the current subband k is greater than the codebook switching overhead, the current Subband k is encoded with a codebook of 0; otherwise, it is encoded with the codebook used by the previous encoded subband that is not all 0.

5.5 Repeat step 5.3 until all subbands complete entropy coding.

3. Code stream synthesis: for each coded sub-band, select the codebook determined in step B for encoding, and complete the code stream synthesis process.

Several processing steps in the above steps will be further described below.

a. Calculate the reference bit consumption value

In the embodiment of the present invention, a Huffman codebook is used to calculate the reference bit consumption value of each coding subband. There are 11 Huffman codebooks in AAC, and their specifications are shown in Table 1. In addition, there is a 0-numbered book used to indicate the codebook used when the quantized spectrum is all 0. The so-called 0-numbered book is actually a virtual codebook , which is used to indicate that the subband is not encoded. Calculate the bit consumption of the quantized spectrum, the calculation process is: first obtain the maximum absolute value t _max of the integer quantized spectrum in the sub-band, and sequentially form the four-dimensional vector [t ₁ t ₂ t ₃ t ₄ ] as an index to search, and accumulate to obtain the total number of code lengths required for entropy encoding of different codebooks. Among them, the index of the signed codebook should be added with the maximum absolute value as the offset, while the index of the unsigned codebook should be added on 1. The search method is:

codeLen1＝Huf_Len1[t1+1][t2+1][t3+1][t4+1]

codeLen2＝Huf_Len2[t1+1][t2+1][t3+1][t4+1]

codeLen5＝Huf_Len5[t1+4][t2+4]+Huf_Len5[t3+4][t4+4]

codeLen6＝Huf_Len6[t1+4][t2+4]+Huf_Len6[t3+4][t4+4]

t1=|t1|, t2=|t2|, t3=|t3|, t4=|t4|

signLen=! ! t1+! ! t2+! ! t3+! ! t4

codeLen3＝Huf_Len3[t1][t2][t3][t4]+signLen (1)

codeLen4=Huf_Len4[t1][t2][t3][t4]+signLen

codeLen7=Huf_Len7[t1][t2]+Huf_Len7[t3][t4]+signLen

codeLen8＝Huf_Len8[t1][t2]+Huf_Len8[t3][t4]+signLen

codeLen9＝Huf_Len9[t1][t2]+Huf_Len9[t3][t4]+signLen

codeLen10＝Huf_Len10[t1][t2]+Huf_Len10[t3][t4]+signLen

codeLen11=Huf_Len11[t1][t2]+Huf_Len11[t3][t4]+signLen

! ! The operation principle of the operator is if t ¹ 0, ! ! t=1, if t=0, ! ! t=0;

The process of retrieving the codebook is judged according to t _max as follows:

Overflow protection mechanism, when t _max ³ 16, first replace the integer spectrum t greater than 16 with 16 as the index for retrieval, and then add escape sequences for the index greater than or equal to 16 after the Huffman codeword. An escape sequence consists of the following three parts:

[escape_prefix][escape_separator][escape_word]

Where [escape_prefix] is N '1', [escape_separator] is a '0', [escape_word] is an unsigned integer of N+4 bits, t satisfies t=2 ^N+4 +[escape_word], increase The number of bits is 2int[log2(t+1)+1]-5.

The nth sub-band integer quantization spectrum obtained by searching the code length uses the bit consumption reference value of the m number book, and records it in the reference bit consumption matrix refBits(m, n), and records the uncalculated one as 65535. The reference bit consumption value of the current subband is

sfbbit(n)=sfLen(n)+min(refBits(m,n)), m=1, 2L, 11(2)

b. Zero subband jump avoidance

In order to avoid multiple codebook serial number switching in subbands that are not encoded and increase additional overhead, when the quantization spectrum encoding of the current subband is not required, the reference bit consumption value of the 11 codebooks in the column corresponding to refBits is bw _sfb (k)×[0.250.750.2510.520.52.50.532], the above values are obtained by looking up the code table, and record bookIndex(k)=0 in the reference codebook index bookIndex, and the reference codebook indexes of the other subbands are all 12 . As shown in Table 2, the corresponding example of the 12th and 13th subbands is 8×[0.250.750.2510.520.52.50.532], where 8 is the bandwidth value of the 12th and 13th subbands.

c. Preprocessing of reference bit consumption value

Do preprocessing on refBits, if the refBits unit is isolated backwards (65535 in the back column), add 9 to the unit (the short block code is 7, and the long and short blocks refer to the length of the AAC processing window). Take Table 2 as an example, rows 1, 2, 3, and 4 in column 2, rows 3 and 4 in column 10, rows 5, 6, 7, and 8 in column 3, and rows 5, 6, 7, and 8 in column 8 Belongs to the backward isolated unit. The processed refBits are shown in Table 3. The row numbers corresponding to the blackened values in the table are the corresponding serial numbers of the codebook with the least reference bit consumption value when each sub-band is coded using codebooks with different serial numbers.

Table 2 Bit overhead number of codebook

Table 3 Bit overhead number of codebook

d. Merge of first-order backtracking codebooks

After preprocessing, starting from the first sub-band, the codebook index is the row number of the minimum value (if there are the same, get the larger row number), i.e. BookIndex(1)=m, refBits[m, 1]=min( refBits[11 to1, 1]), as shown in Table 3, BookIndex(1)=6.

As mentioned above, rule 5.3.1 is applied to sub-band C2, BookIndex(2)=6, rule 5.3.2.2 is applied to sub-band C3, BookIndex(3)=6, rule 5.3.2.2 is applied to sub-band C4, BookIndex( 4)=10, and BookIndex(3) is revised into 10, C5 application rule 5.3.1,, BookIndex(5)=10, for subband 12,13, according to step 5.4, encode BookIndex(12 with 0 number book )=0, BookIndex(13)=0.

After the codebook used by each coded subband is determined according to the above steps, entropy coding is performed on each coded subband to complete code stream synthesis.

Based on the first-order backtracking technology, the present invention uses a simple and efficient codebook merging method to obtain the codebook serial number of the optimized subband, thereby reducing the bit overhead during entropy coding, improving the effect of codestream synthesis, and further improving the overall audio quality. The encoding efficiency of the encoder.

The effectiveness of the present invention is verified with a Megal AAC encoder as a test platform, and the Megal AAC encoder is an MPEG-4AAC encoder implemented by applying the code stream synthesis method of the present invention. The test is performed on 8 stereo clips sampled at 44100Hz and quantized by 16 bits. Each clip lasts for 11.63 seconds. The clips are shown in Table 3. The average value of the 8 clips is taken for each test result. Tests were performed on a workstation with an Intel E6300 processor clocked at 1.86GHz. The average bit rate of the encoding test is set to 80kbps, 128kbps, 192kbps, 256kbps, a total of 4 groups. Respectively adopt all selected minimum values to merge, force 3 sub-bands to merge, force 6 sub-bands to merge, force 9 sub-bands to merge and the codebook merge of this embodiment to calculate the entropy coding consumption value of each frame respectively, wherein provide to the codebook merge The reference bit consumption values of the algorithms are the same.

Table 4 Test Clips

Figures 2, 3, 4, and 5 are the first 100 frames of clip 1 respectively encoded at 80kbps, 128kbps, 192kbps, and 256kbps bit rates, 3 subbands are forced to merge, 6 subbands are forced to merge, 9 subbands are forced to merge and this method Divide the ratio of each frame consumed by these four codebook merging algorithms relative to the minimum value merging bits. Fig. 6 shows that when these 8 clips are encoded at 80kbps, 128kbps, 192kbps, and 256kbps code rates, 3 sub-bands are forced to merge, 6 sub-bands are forced to merge, 9 sub-bands are forced to merge and these four codebook merger algorithms in this embodiment are compared. The average value of the ratio of the bit consumption to the minimum value. From the experimental data, it can be obtained that this embodiment improves the efficiency of the encoder by 10% to 15% compared with various codebook merging algorithms, which is especially obvious at low and medium code rates.

Claims (7)

1. the code stream synthesizing method based on advanced audio coder is characterized in that, comprises following steps:

A, quantization encoding: code stream to be encoded is carried out quantization encoding handle, obtain a plurality of coding subbands;

B, code book merge: the reference bits consumption figures when calculating the subband of respectively encoding and using different code book, the code book that definite subband of respectively encoding uses: press coded sequence, for first coding subband, the code book of selecting reference bits consumption figures minimum in each code book is as using code book; For each the coding subband after first coding subband, if the quantized spectrum of present encoding subband is not 0 entirely, select to make in each code book a present encoding subband and a last coding subband always with reference to the code book of bit consumption value minimum as using code book, specifically comprise: obtain the reference bits consumption figures that the present encoding subband uses each code book, if the code book of reference bits consumption figures minimum is identical with the code book that a last coding subband uses, the present encoding subband uses this code book, otherwise, always always select to use code book by what following three kinds of modes were calculated a present encoding subband and a last coding subband respectively with reference to the mode of bit consumption value minimum with reference to the bit consumption value and according to what make that present encoding subband and last encodes subband:

1) get a coding subband the code selection book always equal the reference bits consumption figures sum of two subbands of encoding as candidate's code book of present encoding subband with reference to the bit consumption value;

2) except that on a coding subband in other all code books the code selection book, selection make two the coding subbands reference bits consumption figures sum minimum code book as the present encoding subband and last one the coding subband common candidate's code book, always equal the reference bits consumption figures sum of two coding subbands with reference to the bit consumption value, if what a last coding subband and a last coding subband used is same code book, always the reference bits consumption figures sum that equals two subbands of encoding with reference to the bit consumption value adds the code book handover overhead.

3) code book of selecting present encoding subband reference bits consumption figures minimum always equals the last one reference bits consumption figures sum of encoding subband and present encoding subband with reference to the bit consumption value and adds the code book handover overhead as candidate's code book;

If the quantized spectrum of a coding subband or a plurality of continuous programming code subbands is 0 entirely, add up reference bits consumption figures or reference bits consumption figures sum that a plurality of continuous programming code subbands of this coding subband or this adopt the previous code book that is not 0 coding subband uses entirely to encode, if this reference bits consumption figures or reference bits consumption figures sum are greater than the code book handover overhead, described coding subband or a plurality of continuous programming code subband use 0 number book coding; Otherwise using previous is not the code book coding that 0 coding subband uses entirely.

C, code stream synthesize: to each coding subband, the use code book of selecting to determine among the step B is encoded, and finishes the synthetic processing of code stream.

2. code stream synthesizing method as claimed in claim 1, it is characterized in that, described reference bits consumption figures is pre-stored among the two-dimensional array refBits, refBits (m, n) be the element value of the capable n row of m among the array refBits, it represents the reference bits consumption figures of the integer quantisation spectrum use m number book of n subband, and (m, initial value n) are changed to maximum 65535 to refBits.

3. code stream synthesizing method as claimed in claim 2 is characterized in that, if the integer quantisation of present encoding subband spectrum is 0 entirely, the code book reference bits consumption figures of this subband correspondence be refBits (m, n)=bw _Sfb(n) * CW ₀(m), wherein, bw _SfbAnd CW ₀Be one-dimension array, bw _Sfb(n) be array bw _SfbIn n element value, the bandwidth of its presentation code subband n, CW ₀(m) be array CW ₀In m element value, its expression is composed the every spectral line bit number that is consumed with the integer quantisation of code book m coding 0 value.

4. code stream synthesizing method as claimed in claim 3 is characterized in that, also comprises the array pre-service, described array pre-service utilizes following formula to carry out: if refBits (m, n+1)=65535, then refBits (m, n)=refBits (m, n)+C, wherein (m n) is the reference bits consumption figures of present encoding subband, refBits (m to refBits, n+1) be the reference bits consumption figures of next coding subband, C is the code book handover overhead.

5. code stream synthesizing method as claimed in claim 4 is characterized in that, described one-dimension array CW ₀=[0.25 0.75 0.251 0.5 2 0.5 2.5 0.5 3 2].

6. code stream synthesizing method as claimed in claim 1 is characterized in that described code book is the Huffman code book, comprises the code book of 11 sequence numbers.

7. code stream synthesizing method as claimed in claim 6 is characterized in that, described reference bits consumption figures is to be that the described Huffman code book of indexed search obtains by the integer quantisation spectrum with the coding subband.

Images