JP3291948B2 - High-efficiency encoding method and apparatus, and transmission medium - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a highly efficient encoding method and apparatus for bit-compressing a digital signal such as a digital audio signal, and a transmission medium (including a recording medium) for transmitting the compressed data. The present invention relates to a high-efficiency encoding method and apparatus that changes the bit allocation amount of a processing block according to the shape of energy for each, and a transmission medium.

[0002]

2. Description of the Related Art There are various methods and apparatuses for high-efficiency encoding of signals such as audio or voice. For example, a time domain audio signal or the like is divided into blocks for each unit time, and a time axis signal for each block is obtained. Is converted into a signal on the frequency axis (orthogonal transform), divided into a plurality of frequency bands, and a so-called transform coding method, which is a block frequency band division method for encoding each band, an audio signal in the time domain, and the like. Is a non-blocking frequency band division scheme (sub-band coding: S
BC) system and the like. Further, a high-efficiency coding method and apparatus combining the above-mentioned band division coding and transform coding are also considered. In this case, for example, after performing band division by the above-mentioned band division coding method, Then, the signal for each band is orthogonally transformed into a signal in the frequency domain by the above-described transform coding method, and encoding is performed for each band subjected to the orthogonal transformation.

Here, as a band division filter used in the above-mentioned band division coding system, for example, QMF (Q
There is a filter such as the uadrature Mirror filter, which is described, for example, in the document "Digital coding of speech in subbands".
n subbands "RECrochiere, Bell Syst.Tech. J., Vo
l.55, No.8 1976). This QMF filter divides a band into two equal bandwidths, and is characterized in that so-called aliasing does not occur when the divided bands are combined later. In addition, the document "Polyphase Quadratic
Filters-A New Band Division Coding Technique "(" Polyphase
Quadrature filters -A new subband coding techniqu
e ", Joseph H. Rothweiler ICASSP 83, BOSTON)
An equal bandwidth filter splitting technique is described. This polyphase quadrature filter is characterized in that a signal can be divided at a time when divided into a plurality of bands of equal bandwidth.

As the above-mentioned orthogonal transform, for example, an input audio signal is divided into blocks in a predetermined unit time (frame), and a fast Fourier transform (F
FT), a discrete cosine transform (DCT), a modified DCT transform (MDCT), and the like, to transform the time axis to the frequency axis. This MDCT
For more information, see the document "Subband / Transform Coding Using Fil," which uses filter bank design based on time-domain aliasing cancellation.
ter Bank Designs Based on Time Domain Aliasing Can
cellation, "JPPrincen ABBradley, Univ. of Surr
ey Royal Melbourne Inst. of Tech. ICASSP 1987).

Further, as a frequency division width when quantizing each frequency component divided into frequency bands, for example, there is a band division in consideration of human auditory characteristics. That is, an audio signal may be divided into a plurality of bands (for example, 25 bands) with a bandwidth such that a higher band generally called a critical band (critical band) has a wider bandwidth. When encoding data for each band at this time, predetermined bits are allocated to each band, or encoding is performed by adaptive bit allocation to each band. For example, when the MDCT coefficient data obtained by the MDCT processing is encoded by the bit allocation, the MDCT coefficient for each band obtained by the MDCT processing for each block is used.
Coding is performed on the coefficient data with an adaptive number of allocated bits.

The following two methods and devices are known as the above-mentioned bit allocation method and a device therefor. For example,
Reference "Adaptive Transf Coding of Speech Signal"("Adaptive Transf
ormCoding of Speech Signals ", IEEE Transactions of
Accoustics, Speech, andSignal Processing, vol.ASS
In P-25, No. 4, August 1977), bits are allocated based on the magnitude of the signal for each band. In addition, for example, the document “Critical band encoder—Digital encoding for perceptual requirements of auditory system” (“The critical band cod
er --digital encoding of the perceptual requireme
nts of the auditory system ", MAKransner MIT, ICA
SSP 1980) describes a method and apparatus for obtaining a required signal-to-noise ratio for each band and performing fixed bit allocation by using auditory masking.

[0007]

In the above-described conventional high-efficiency coding method and apparatus, when the adaptive bit allocation amount is calculated for each band, the total number of bits of all the quantized blocks is fixed. The number of bits is unlikely to be equal to the bit rate of the bit rate, and in most cases, it is necessary to adjust the bit fraction to make the number of bits equal to the predetermined bit rate. In this bit fraction adjustment, for example, the bit fraction adjustment is simply performed according to the priority order depending on the frequency, and therefore, the bit fraction adjustment more suitable for the input signal cannot be performed. Therefore, particularly when the bit rate is low and the number of usable bits is small, for example, good sound quality cannot be obtained in terms of audibility.

[0008] The present invention has been made in view of such circumstances, and provides an efficient and highly efficient encoding method and apparatus utilizing the auditory characteristics, and a transmission medium.
The purpose is to prevent sound quality deterioration at a low bit rate and to improve sound quality at the same bit rate.

[0009]

SUMMARY OF THE INVENTION A high efficiency coding method and apparatus according to the present invention decomposes an input signal into a plurality of frequency band components and divides the frequency band components with respect to time and frequency into a plurality of two-dimensional blocks. In each case, bit allocation information required for quantization is obtained as a decimal value according to the frequency band component, and the obtained bit allocation information is rounded to calculate a bit allocation amount of an integer value for each two-dimensional block, Each of the above 2
The bit allocation amount calculated for each dimensional block is adjusted to be equal to the bit amount determined by the predetermined bit rate, the bit allocation amount for each two-dimensional block is determined, and according to the determined bit allocation amount, The frequency band component is quantized for each of the two-dimensional blocks, and when determining the bit allocation amount, the total bit allocation amount calculated for each of the two-dimensional blocks is determined by a predetermined bit rate. If not equal to the above, the above object is achieved by adjusting the bit allocation amount based on the information of the rounding process in the bit allocation amount calculation step.

In the high-efficiency encoding method and apparatus according to the present invention, when determining the bit allocation amount,
When the total bit allocation amount calculated for each dimension block is larger than the bit amount determined by the predetermined bit rate, the number of bits corresponding to the two-dimensional block rounded up in the bit allocation amount calculation step is preferentially reduced. . Also, when determining the bit allocation amount, the above 2
When the sum of the bit allocation amounts calculated for each dimension block is smaller than the bit amount determined by the default bit rate,
The number of bits corresponding to the two-dimensional block that has been truncated in the bit allocation amount calculation step is preferentially increased.

Further, in the high-efficiency encoding method and apparatus of the present invention, information on round-up and / or round-down processing when calculating the bit allocation amount is used as the information on the rounding processing. In addition, as the information of the rounding processing, a numerical value of a digit after the decimal point of the bit allocation information when calculating the bit allocation amount is also used.

A transmission medium according to the present invention is a transmission medium for transmitting an input signal compressed and encoded by a high-efficiency encoding method, wherein the input signal is decomposed into a plurality of frequency band components, and For each of a plurality of two-dimensional blocks obtained by dividing the component with respect to time and frequency, bit allocation information necessary for quantization is obtained as a decimal value in accordance with the frequency band component, and the obtained bit allocation information is rounded to obtain 2 bits. The bit allocation amount of the integer value for each dimension block is calculated, and
The bit allocation amount calculated for each two-dimensional block is determined by adjusting the bit allocation amount calculated for each dimensional block to be equal to the bit amount determined by the predetermined bit rate, and according to the determined bit allocation amount. When performing quantization of the frequency band component for each of the two-dimensional blocks, if the sum of the bit allocation amounts calculated for each of the two-dimensional blocks is not equal to the bit amount determined by a predetermined bit rate, Based on the information of the rounding process performed at the time of calculation, by adjusting the bit allocation amount and transmitting a signal that has been compression-encoded by a convergence efficiency coding method of determining the bit allocation amount of each two-dimensional block,
Achieve the above objectives.

[0013]

[0014]

[0015]

According to the present invention, when converting bit information having a decimal part calculated for each two-dimensional block into an integer, the fraction of the bits is adjusted in consideration of the information of the rounding processing to be performed. By determining the bit allocation of the dimension block,
Efficient bit allocation can be performed by making better use of the auditory characteristics, and efficient compression can be performed without lowering the block floating efficiency.

As described above, it is possible to obtain better sound quality at the same bit rate. Also,
In order to obtain the same sound quality, it can be performed at a lower bit rate.

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

In this embodiment, an input digital signal such as an audio PCM signal is subjected to band division coding (SBC), adaptive conversion coding (ATC), and adaptive bit allocation (APC).
-Efficient coding is performed using each of the techniques of AB). This technique will be described with reference to FIG.

In a specific high-efficiency encoding apparatus to which the high-efficiency encoding method of the present invention shown in FIG. 1 is applied, an input digital signal is divided into a plurality of frequency bands by a filter or the like, and each frequency band is divided into a plurality of frequency bands. The orthogonal transform is performed for each band, and the obtained spectrum data on the frequency axis is adaptively bit-distributed and encoded for each so-called critical bandwidth (critical band) in consideration of human auditory characteristics described later. At this time, in the high band, a band obtained by further dividing the critical bandwidth is used. Of course, a non-blocking frequency division width by a filter or the like may be an equal division width.

Furthermore, in the embodiment of the present invention, the block size (orthogonal transform block length) is adaptively changed according to the input signal before the orthogonal transform, and the critical bandwidth (the critical bandwidth in the critical band unit or high band) is changed. Floating processing is performed on blocks (two-dimensional blocks) obtained by further subdividing critical bands. Note that this critical band is a frequency band divided in consideration of human auditory characteristics, and the narrow band noise of the same intensity near the frequency of a certain pure tone masks the noise when the pure tone is masked. It is the band that you have. In this critical band, the higher the frequency band, the wider the bandwidth. For example, the entire frequency band of 0 to 20 kHz is 25, for example.
Has been divided into critical bands.

That is, in FIG. 1, an audio PCM signal of, for example, 0 to 22 kHz is supplied to the input terminal 10. This input signal is divided into a band of 0 to 11 kHz and a band of 11 to 22 kHz by a band division filter 11 such as a so-called QMF, for example.
The band signals are also divided by a band division filter 12, such as a so-called QMF, into a band of 0 to 5.5 kHz and a band of 5.5 to 11 k.
Hz band.

11k from the band division filter 11
The signal in the 22 kHz band is represented by M which is an example of an orthogonal transformation circuit.
DCT (Modified Discrete Cosine Transform) circuit 1
5.5k from the band division filter 12
The signal in the ~ 11 kHz band is sent to the MDCT circuit 14,
The signals in the 0 to 5.5 kHz band from the band division filter 12 are sent to the MDCT circuit 15 to be subjected to the MDCT processing. Note that each of the MDCT circuits 13 to
In the block 15, the block decision circuits 19, 2 provided for each band are used.
M based on the block size determined by 0,21
DCT processing is performed.

Here, the above-mentioned block decision circuits 19 to 21
FIGS. 2A and 2B show specific examples of the block size in each of the MDCT circuits 13 to 15 determined by the equation (1). FIG. 2A shows a case where the orthogonal transform block size is long (orthogonal transform block size in the long mode).
FIG. 2B shows a case where the orthogonal transform block size is short (orthogonal transform block size in the short mode).
Is shown.

In the specific example of FIG. 2, each of the three filter outputs has two orthogonal transform block sizes. That is, for a signal in the 0-5.5 kHz band on the low band side and a signal in the 5.5 kHz-11 kHz band in the middle band, in the case of a long block length ((A) in FIG. The number is 128 samples, and when a short block is selected (FIG. 2B), the number of samples in one block is a block for every 32 samples. On the other hand, for a signal in the 11 kHz to 22 kHz band on the high frequency side, in the case of a long block length ((A) in FIG. 2), the number of samples in one block is 256, and a short block is selected. In the case ((B) of FIG. 2), the number of samples in one block is a block for every 32 samples. When a short block is selected in this way, the number of samples of the orthogonal transform block in each band is kept the same, and the higher the frequency, the higher the time resolution and the type of window used for blocking.

Referring again to FIG. 1, the information indicating the block size determined by the block determination circuits 19 to 21 is as follows.
The signals are sent to adaptive bit assignment encoding circuits 16, 17, and 18 to be described later, and are also output from output terminals 23, 25, and 27.

The spectrum data or the MDCT coefficient data in the frequency domain obtained by the MDCT processing in each of the MDCT circuits 13 to 15 are grouped into a so-called critical band (critical band) or a band obtained by further dividing the critical band in a high band. And sent to the adaptive bit allocation encoding circuits 16-18.

In the adaptive bit allocation coding circuits 16 to 18, each spectrum is determined according to the information on the block size and the number of bits allocated to each critical band (critical band) or the critical band in the high band. The data (or MDCT coefficient data) is requantized (normalized and quantized).

Each of these adaptive bit allocation encoding circuits 16 to
The data encoded by 18 is output to output terminals 22, 2
It is taken out via 4, 26. The adaptive bit allocation encoding circuits 16 to 18 also include a scale factor indicating what kind of signal size has been normalized and bit length information indicating what kind of bit length has been used for quantization. At the same time, the output terminal 22,
24 and 26.

Each of the MDCT circuits 13 to 13 in FIG.
From the output of No. 15, the energy of each band obtained by further dividing the critical band in the critical band or the high band is calculated, for example, by calculating the root mean square of each amplitude value in the band. Desired. Of course, the scale factor itself may be used for subsequent bit allocation. In this case, a new calculation for energy calculation is not required, so that the hardware scale is saved. Further, instead of the energy for each band, a peak value, an average value, or the like of the amplitude values can be used.

Next, the adaptive bit allocation coding circuits 16-1
8 will be described.

The operation of the adaptive bit allocation encoding circuit in this case will be described with reference to FIG. 3. The magnitude of the MDCT coefficient is obtained for each block, and the MDCT coefficient is supplied to the input terminal 801. MDCT supplied to the input terminal 801
The coefficient is provided to the energy calculation circuit 803 for each band and the smoothness calculation circuit 808 of the spectrum. The energy calculation circuit 803 for each band calculates the signal energy for each band obtained by further subdividing the critical band in the critical band or the high band. The energy for each band calculated by the energy calculation circuit 803 for each band is stored in the energy-dependent bit distribution circuit 8.
04 and the smoothness calculating circuit 808 for the spectrum.

In the energy-dependent bit distribution circuit 804,
Using the energy from the per-band energy calculation circuit 803 and the total available bits from the total available bits generation circuit 802, a bit distribution depending on the energy of the input signal is calculated. Perform using The higher the tonality of the input signal, that is, the greater the unevenness of the spectrum of the input signal, the greater the ratio of this bit amount to the above 128 kbps. In addition,
To detect the unevenness of the input signal, the sum of the absolute values of the differences between the block floating coefficients of adjacent blocks is used as an index. Then, for the determined available bit amount, bit allocation is performed in proportion to the logarithmic value of the energy of each band.

The bit allocation calculating circuit 805 depending on the permissible noise level of the hearing first calculates the permissible noise amount for each critical band in consideration of the so-called masking effect and the like based on the spectrum data divided for each critical band. Are allocated by subtracting the energy-dependent bits from the total available bits so as to give the permissible noise spectrum.

The number of energy-dependent bits and the number of bits depending on the permissible noise level obtained in this way are added in an adder 806, and the primary bit distribution circuit 81
5 is supplied.

In the primary bit distribution circuit 815, the adder 8
The bit information having a decimal part corresponding to each of the two-dimensional blocks starting from 06 (hereinafter, the two-dimensional block for quantization is particularly referred to as a quantization block) is converted into an integer bit value, and the temporary bit allocation is performed. Do.

The bit information at the input of the adder 806 of the present embodiment, in other words, the output of the multipliers 811 and 812 can have a decimal part. This allows
It is possible to obtain the bit allocation amount for the corresponding quantized block more precisely.

A specific description will be given with reference to FIG. One
Focusing on only one quantization block, for example, the output of the energy-dependent bit allocation circuit 804 is 2 bits, the output of the bit allocation circuit 805 depending on the permissible noise level is 1 bit, and among the outputs of the bit division ratio determination circuit 809, The output to the energy dependent bit distribution circuit 804 is 0.
7. If the output to the bit allocation circuit 805 depending on the permissible noise level is 0.3, the output of the multiplier 811 is 1.4 bits and the output of the multiplier 812 is 0.3 bits. . When the values of the decimal part are held and added by the adder 806 as in the present embodiment, the output of the adder 806 becomes 1.7 bits, and thereafter, the primary bit distribution circuit 815
Is converted to an integer, and the primary bit allocation amount for this quantized block becomes 2 bits. For example, if the input of the adder 806 has no decimal part, the outputs of the multipliers 811 and 812 are converted to integers by rounding off the value of the first decimal point, for example. For example, in the above example, 1.4 bits of the output of the multiplier 811 are converted into 1 bit, and 1.7 bits of the output of the multiplier 812 are converted into 2 bits. When these integers are added by the adder 806, the output of the adder 806, that is, the primary bit allocation amount becomes 3 bits. As is clear from comparison of the above examples, by determining the primary bit allocation amount while retaining the decimal part, the bit allocation amount adapted to the input signal can be calculated more precisely.

If the bit information at the input of the adder 806 does not have a decimal part, the amount of calculation in each multiplier and adder is reduced, and the outputs and additions of the multipliers 811 and 812 are reduced. Since the width of the terminal and the transmission line at the input of the device 806 can be reduced, the hardware scale can be reduced.

As described above, in the primary bit distribution circuit 815, the bit information for each quantized block holding the decimal part is converted into an integer. Referring to FIG. 4, bit information for each quantized block having a decimal part indicated by L1 in the figure is rounded to an integer value, and is converted into an integer bit number indicated by L2 in the figure. For example, in the quantization block B1 in the figure, the bit information of 3.7 bits is rounded to an integer of 4 bits by rounding off the first decimal place. Note that the rounding process in FIG. 4 uses rounding to minimize the rounding error. However, different rounding processes such as changing the threshold value, rounding down 3 or less, and rounding up 4 or more may be performed. It is possible.

The primary bit distribution circuit 815 outputs to the memory 816 rounding information indicating whether rounding up or rounding down has been performed in the rounding process to the integer.

The memory 816 includes a primary bit distribution circuit 81
5 temporarily holds rounding information indicating either rounding up or rounding down for each quantization block, and outputs the rounding information to the bit fraction adjusting circuit 814. The memory 816 has a capacity of one bit for one quantization block. For example, in the case of FIG. 4, since there are eight quantization blocks, the memory 816 has a capacity of 8 bits.

The bit fraction adjusting circuit 814 uses the number of bits for each quantized block from the primary bit distribution circuit 815, the rounding information from the memory 816, and the total available bits from the total available bit generation circuit 802. ,
In this embodiment, the fraction of bits is adjusted so as to match the bit rate of 128 kbps. Primary bit distribution circuit 8
Since the total number of bits used, ie, the bit rate, at the time of outputting 15 is the number of bits calculated for each quantization block, in most cases, the total available bit generation circuit 80
2 does not match the total number of available bits output.
Therefore, a predetermined bit rate, 128 kb in this embodiment.
To match ps, a bit fraction adjustment circuit 814 is required.

For example, when the output of the primary bit distribution circuit 815 has a number of bits exceeding the predetermined bit rate, the band having the least influence on the sound quality from the viewpoint of the audibility is, for example, shown in FIG. As shown in FIG. 5, a method of sequentially decreasing one bit at a time from the highest frequency block to the low frequency side, or reducing the number of allocated bits from the highest frequency block to 0 as shown in FIG. That is, there is a method of performing band limitation. 5A and 5B, four bits are reduced for each of FIGS. 5A and 5B, and the hatched portions indicate the reduced bits.

The bit fraction adjusting circuit 814 of this embodiment uses the rounding information for each quantized block from the memory 816 and preferentially goes from the high-frequency side to the low-frequency side for the rounded-up quantized block. The fraction is adjusted so that the bit rate becomes equal to 128 kbps by using the method of reducing one bit at a time. Referring to FIG. 6, FIG.
Similarly, L1 in FIG. 6 indicates bit information having a decimal part input to the primary bit allocation circuit 815, and L2 indicates the number of bits obtained by converting the bit information into an integer by rounding in the primary bit allocation circuit 815. Is shown. As is clear from FIG. 6, the quantized blocks rounded up by the rounding process are blocks B3, B5, B6, and B8, and the quantized blocks rounded down are blocks B1, B2, B4, and B7. Information about these roundings is recorded in the memory 816. The bit fraction adjustment circuit 814 uses the rounding information from the memory 816 to sequentially reduce the rounded-up quantized block by one bit from the high-frequency side to the low-frequency side.

A specific bit reduction method will be described with reference to FIG. In FIG. 6, as in FIG. 5, it is assumed that the number of bits becomes equal to the predetermined bit rate by reducing 4 bits.

First, focusing on the block B8 on the highest frequency side, it can be seen that rounding up has been performed based on the rounding information from the memory 816, so that one bit is reduced. Next, focusing on the adjacent low-frequency side quantization block B7, since the truncation has been performed, the bit is not reduced. Next, paying attention to the adjacent low-frequency-side quantization block B6, since the rounding has been performed, a reduction of one bit is performed. This series of processing is repeated until the number of bits becomes equal to the predetermined bit rate. As a result, a portion where the number of bits is reduced is indicated by hatching in FIG. In this process, for example, in this process, the number of bits is sequentially reduced from the high frequency side to the low frequency side, and reaches the quantization block B1 on the lowest frequency side. Returning to the quantization block B8 on the highest frequency side, bit reduction is performed by the method shown in FIG. 5 (A) or (B).

Conversely, for example, the primary bit fraction adjustment circuit 81
In the case where the output of bit No. 4 has a number of bits lower than the predetermined bit rate, the band that has the greatest audibility to the sound quality, for example, as shown in FIG. To increase the number of bits one bit at a time. In FIG. 7, the number of bits is increased by 4 bits, and the shaded portions indicate the increased bits.

The bit fraction adjusting circuit 814 of this embodiment uses the rounding information for each quantized block from the memory 816, and preferentially goes from the low frequency side to the high frequency side with respect to the truncated quantized block. The fraction of bits is adjusted so that the bit rate becomes equal to 128 kbps by using a method of increasing one bit at a time. Referring to FIG. 8, as in FIG. 4, L1 in FIG.
5 indicates bit information having a decimal part, and L2 indicates the number of bits obtained by converting the bit information into an integer by rounding in the primary bit distribution circuit 815. As in FIG. 6, the quantized blocks that have been truncated by the rounding process are blocks B1, B2, B4, and B7. In the bit fraction adjusting circuit 814 of the present embodiment, these truncated quantized blocks are preferentially incremented one bit at a time from the low band to the high band.

More specifically, in FIG. 8, as in FIG. 7, it is assumed that the number of bits becomes equal to the predetermined bit rate by an increase of 4 bits. Focusing on the quantization block B1 on the lowest frequency side, it can be seen that truncation is performed by the rounding information from the memory 816.
Increment by bits. Next, paying attention to the adjacent high-frequency side quantization block B2, since the truncation has been performed, an increase of one bit is performed. Next, focusing on the adjacent high-frequency quantization block B3, the bits are not increased because the rounding has been performed. This series of processing is repeated until the number of bits becomes equal to the predetermined bit rate. As a result, a portion where the number of bits is increased is indicated by hatching in FIG. In this processing, for example, in this process, the number of bits is sequentially increased from the low-frequency side to the high-frequency side, and reaches the quantization block B8 on the highest frequency side. Lowest quantization block B1
Then, the number of bits is increased by the method shown in FIG.

As described above, the rounding information in the integer conversion processing in the primary bit distribution circuit 815 is used for adjusting the bit fraction to match the predetermined bit rate in the bit fraction adjustment circuit 814, so that the input signal , Bit allocation can be performed more adaptively according to the characteristics described above, and it is possible to obtain good sound quality in terms of audibility. In particular, when a low bit rate is used, the effect on the sound quality of one bit increases, which is effective.

In the memory 816, instead of the rounding information indicating either rounding up or rounding down, the value of the decimal part of the output of the adder 806, specifically, for example, the value of the first digit of the decimal point is stored. It is also conceivable to retain the data. The numerical value indicates, for example, 7 when the output of the adder 806 for a certain quantization block is 1.7 bits.
In this method, the capacity of the memory 816 increases,
Since the required number of bits per quantization block is 4 bits, for example, in the case of FIG. 4, since there are eight quantization blocks, the memory 816 has a capacity of 32 bits.

The value of the decimal part held by the memory 816 is sent to the bit fraction adjusting circuit 814, and is used to adjust the number of bits to the same as the predetermined bit rate. For example, when the output of the primary bit distribution circuit 815 has a number of bits exceeding a predetermined bit rate, the method described with reference to FIG. Bit reduction was performed based on the priority order depending on the frequency to the side, but in the method using the numerical value of the decimal point, the bit reduction is performed using the priority order according to the magnitude of the error due to the round-up processing. . For example, if there are two quantized blocks that have been rounded up and rounded up to 2 bits and the outputs of the adders 806 are 1.5 and 1.9, respectively, the error generated by the rounding up process is 0. 0.5 and 0.1, and the former has a larger error. This means that 0.5 bits extra than the originally required number of bits are allocated.
Therefore, when performing bit reduction, it is better to preferentially reduce the number of bits for the former quantization block having a large error,
Efficient bit allocation more suitable for the input signal can be performed.

Conversely, for example, when the output of the primary bit distribution circuit 815 has a number of bits lower than the predetermined bit rate, the method described with reference to FIG. From the frequency side to the high frequency side, the bit was increased based on the priority depending on the frequency,
In the method using the numerical value of the decimal point, the number of bits is increased using a priority order according to the magnitude of an error caused by the truncation processing. For example, if there are two quantized blocks that have been rounded down and rounded down to 2 bits and the outputs of the adders 806 are 2.1 and 2.4, respectively, the errors generated by the rounding down are 0. .
1, 0.4, the latter having a larger error. This means that the bits are distributed by 0.4 bits less than the bit number originally considered necessary. Therefore, when increasing bits, it is possible to perform efficient bit allocation more suitable for an input signal by preferentially increasing bits of the latter quantization block having a large error.

The bits adjusted to a bit rate of, for example, 128 kbps are divided into a plurality of critical bands in each critical band or in a high band by the adaptive bit allocation coding circuits 16 to 18 in FIG. Each spectrum data (or MDCT coefficient data) is re-quantized according to the number of bits allocated to the assigned band. The data encoded in this way is output to the output terminals 22, 24,
Retrieved via 26.

More specifically, the permissible noise spectrum calculation circuit in the permissible noise spectrum dependent bit allocation circuit 805 will be described.
5 is assigned to the bit distribution circuit 805.
To the permissible noise spectrum calculation circuit.

FIG. 9 is a block circuit diagram showing a schematic configuration of one specific example in which the allowable noise spectrum calculation circuit is collectively described. In FIG. 9, an input terminal 521 includes:
Frequency domain spectrum data is supplied from the MDCT circuits 13 to 15.

The input data in the frequency domain is sent to the energy calculation circuit 522 for each band, and the energy for each critical band (critical band) is calculated, for example, by summing the square of each amplitude value within the band. Required. Instead of the energy for each band, a peak value or an average value of the amplitude value may be used. As an output from the energy calculation circuit 522, for example, the spectrum of the sum value of each band is generally called a bark spectrum. FIG. 10 shows such a bark spectrum SB for each critical band. However, in FIG. 10, the number of the critical bands is represented by 12 bands (B1 to B12) to simplify the illustration.

Here, in order to consider the influence of the bark spectrum SB on so-called masking, a convolution process is performed such that the bark spectrum SB is multiplied by a predetermined weighting function and added. Therefore, the output of the energy calculation circuit 522 for each band, that is, each value of the bark spectrum SB is sent to the convolution filter circuit 523. The convolution filter circuit 523
Includes, for example, a plurality of delay elements for sequentially delaying input data, a plurality of multipliers (for example, 25 multipliers corresponding to each band) for multiplying an output from these delay elements by a filter coefficient (weighting function). , And a sum adder for summing the outputs of the multipliers.

The above-mentioned masking refers to a phenomenon in which a certain signal causes another signal to be masked and become inaudible due to human auditory characteristics. There are an axis masking effect and a simultaneous masking effect by a frequency domain signal. Due to these masking effects, even if there is noise in the masked portion, this noise will not be heard. For this reason, in an actual audio signal, noise within the masked range is regarded as acceptable noise.

A specific example of the multiplication coefficient (filter coefficient) of each multiplier of the convolution filter circuit 523 is as follows. When the coefficient of the multiplier M corresponding to an arbitrary band is 1, the multiplier M −1, the coefficient 0.15, the multiplier M-2, the coefficient 0.0019, and the multiplier M-3, the coefficient 0.00000.
086, a coefficient 0.4 by a multiplier M + 1, and a multiplier M + 2
By multiplying the output of each delay element by the coefficient 0.06 by the multiplier M + 3 and the coefficient 0.007 by the multiplier M + 3, the convolution process of the bark spectrum SB is performed. However, M is 1
It is an arbitrary integer of 25.

Next, the output of the convolution filter circuit 523 is sent to a subtractor 524. The subtracter 524 calculates a level α corresponding to an allowable noise level described later in the convolved area. The level α corresponding to the permissible noise level (permissible noise level) is, as described later, a level at which the permissible noise level of each critical band is obtained by performing inverse convolution processing. is there.

Here, the subtractor 524 is supplied with an allowance function (a function expressing a masking level) for obtaining the level α. The level α is controlled by increasing or decreasing the allowable function. The permissible function is a (n-ai) function generation circuit 52 described below.
5.

That is, the level α corresponding to the allowable noise level can be obtained by the following equation, where i is a number sequentially given from the lower band of the critical band.

Α = S− (n−ai) In this equation, n and a are constants and a> 0, and S is the intensity of the convolution-processed Bark spectrum.
i) is the tolerance function. As an example, n = 38, a = −0.
5 can be used.

In this way, the level α is obtained, and this data is transmitted to the divider 526. In the divider 526, the level α in the convolved area is inversely convolved. Therefore, by performing the inverse convolution processing, a masking threshold can be obtained from the level α. That is, this masking threshold becomes the allowable noise spectrum. Note that the above inverse convolution process requires a complicated operation, but in this embodiment, inverse convolution is performed using a simplified divider 526.

Next, the above-mentioned masking threshold is transmitted to the subtractor 528 via the synthesizing circuit 527. Here, the output from the energy detection circuit 522 for each band, that is, the above-described bark spectrum SB is supplied to the subtracter 528 via the delay circuit 529. Therefore, the subtractor 528 performs a subtraction operation between the masking threshold and the bark spectrum SB, and as shown in FIG. 11, the bark spectrum SB becomes the masking threshold M.
The level below the level indicated by the level S is masked. Note that the delay circuit 529 is connected to the synthesis circuit 52.
7 is provided to delay the bark spectrum SB from the energy detection circuit 522 in consideration of the amount of delay in each circuit before 7.

The output from the subtracter 528 is taken out via an allowable noise correction circuit 530 and an output terminal 531. For example, a ROM in which information on the number of bits to be allocated is stored in advance.
Etc. (not shown). This ROM or the like distributes each band in accordance with the output (the level of the difference between the energy of each band and the output of the noise level setting means) obtained from the subtraction circuit 528 via the allowable noise correction circuit 530. Output bit number information.

In this way, the energy-dependent bits and the bits depending on the permissible noise level are added, and the information on the number of allocated bits is added to the adaptive bit allocation coding circuits 16-1.
8, the respective spectral data in the frequency domain from the MDCT circuits 13 to 15 are quantized by the number of bits allocated to each band.

That is, in summary, in the adaptive bit allocation coding circuits 16 to 18, in each band of the critical band (for each critical band) or in a high band, the band is obtained by further dividing the critical band into a plurality of bands. The spectrum data for each band is quantized by the number of bits allocated according to the level of the difference between the energy or the peak value and the output of the noise level setting means.

By the way, at the time of synthesizing by the synthesizing circuit 527, data indicating a so-called minimum audible curve RC, which is a human auditory characteristic as shown in FIG. The above-mentioned masking threshold MS can be synthesized. At this minimum audible curve, if the absolute noise level is below this minimum audible curve, the noise will not be heard. This minimum audible curve will differ depending on the playback volume during playback, for example, even if the coding is the same, but in a realistic digital system, the way in which music enters the 16-bit dynamic range, for example, differs greatly. For example, if quantization noise in the most audible frequency band around 4 kHz is not heard, it is considered that quantization noise below the level of the minimum audible curve is not heard in other frequency bands. Therefore, for example, the dynamic range of the system is 4 k
It is assumed that the noise is not heard around Hz, and the minimum audible curve RC and the masking threshold M
If an allowable noise level is obtained by synthesizing S and S, the allowable noise level in this case can be up to the shaded portion in FIG. In this embodiment, the 4 kHz level of the minimum audible curve is adjusted to the lowest level corresponding to, for example, 20 bits. FIG. 12 also shows the signal spectrum SS.

In the allowable noise correction circuit 530,
The permissible noise level in the output from the subtractor 528 is corrected based on, for example, information on the equal loudness curve sent from the correction information output circuit 533. Here, the equal loudness curve is a characteristic curve relating to human auditory characteristics. For example, the loudness curve is obtained by calculating the sound pressure of sound at each frequency that sounds as loud as the pure tone of 1 kHz, and connecting the curves with each other. Also called a sensitivity curve. This equal loudness curve is the minimum audible curve R shown in FIG.
It draws substantially the same curve as C. In this equal loudness curve, for example, at around 4 kHz, even if the sound pressure falls by 8 to 10 dB below 1 kHz, it sounds as large as 1 kHz, and conversely, at around 50 Hz, the sound pressure must be about 15 dB higher than the sound pressure at 1 kHz. It doesn't sound the same size. For this reason, it can be seen that noise exceeding the level of the minimum audible curve (allowable noise level) preferably has a frequency characteristic given by a curve corresponding to the equal loudness curve. From this, it can be seen that correcting the allowable noise level in consideration of the equal loudness curve is suitable for human auditory characteristics.

The above-described spectral shape depending on the permissible noise level is created by bit allocation using a certain percentage of the total available bits of 128 Kbps. This ratio decreases as the tonality of the input signal increases.

Next, a description will be given of a bit amount division method between the two bit distribution methods.

Returning to FIG. 3, the signal from the input terminal 801 to which the output of the MDCT circuit is supplied is also supplied to a spectrum smoothness calculation circuit 808, where the spectrum smoothness is calculated. In this embodiment, a value obtained by dividing the sum of the absolute values of the differences between the adjacent values of the absolute value of the signal spectrum by the sum of the absolute values of the signal spectrum is calculated as the smoothness of the spectrum.

The above-described spectrum smoothness calculating circuit 808
Is supplied to a bit division ratio determination circuit 809, where the energy-dependent bit allocation and the bit division ratio between the bit allocations based on the permissible noise spectrum are determined. The bit division ratio is calculated by the spectrum smoothness calculation circuit 80.
It is considered that the larger the output value of 8 is, the less smooth the spectrum is, so that bit allocation is performed with more emphasis on bit allocation based on the permissible noise spectrum than on energy-dependent bit allocation. The bit division ratio determination circuit 809 includes a multiplier 8 for controlling the size of the energy-dependent bit allocation and the size of the bit allocation based on the permissible noise spectrum.
Send control output to 11 and 812. Here, if the output of the bit division ratio determination circuit 809 to the multiplier 811 takes a value of 0.8 so that the spectrum is smooth and the energy-dependent bit allocation is weighted, the multiplier 812 The output of the bit division ratio determining circuit 809 is 1-0.8 = 0.2. The outputs of these two multipliers are added in an adder 806 to form final bit allocation information, which is output from an output terminal 807.

FIGS. 13 and 14 show how bits are allocated at this time. FIGS. 15 and 16 show the corresponding quantization noise. FIG. 13 shows a case where the signal spectrum is relatively flat, and FIG. 14 shows a case where the signal spectrum shows high tonality. Also,
In FIGS. 13 and 14, QS indicates a signal level dependent bit amount, and QN indicates a permissible noise level dependent bit allocation bit amount. 15 and 16, L indicates the signal level, NS indicates the noise reduction due to the signal level dependence, and NN indicates the noise reduction due to the permissible noise level-dependent bit allocation. .

First, in FIG. 15 showing a case where the spectrum of the signal is relatively flat, the bit allocation depending on the permissible noise level is useful for obtaining a large signal-to-noise ratio over the entire band. However, relatively low bit allocation is used in the low and high bands. This is because the sensitivity to noise in this band is small acoustically. Although the amount of the bit allocation depending on the signal energy level is small in amount, in this case, the bit allocation is focused on the high-frequency region where the signal level in the middle and low frequencies is high so as to generate a white noise spectrum.

On the other hand, as shown in FIG. 16, when the signal spectrum exhibits a high tonality, the bit allocation amount depending on the signal energy level increases, and the quantization noise decreases with a very narrow band of noise. Used to reduce The concentration of the bit allocation depending on the permissible noise level is less tight.

As shown in FIG. 16, the improvement of the characteristics in the isolated spectrum input signal is achieved by the sum of the two bit allocations.

FIG. 17 shows a basic high efficiency decoding apparatus according to the embodiment of the present invention for decoding again a signal coded by the high efficiency coding apparatus shown in FIG.

In FIG. 17, the quantized MDCT coefficients of each band are input to decoding device input terminals 122 and 124,
The block size information provided to 126 and used is provided to input terminals 123, 125, and 127. The decoding circuits 116, 117 and 118 release the bit allocation using the adaptive bit allocation information.

Next, the IMDCT circuits 113, 114, 1
At 15, the signal in the frequency domain is converted to a signal in the time domain. The time domain signals of these partial bands are decoded by the IQMF circuits 112 and 111 into whole-area signals and sent to the output terminal 110.

Next, the transmission medium of the embodiment of the present invention is for recording or transmitting the signal encoded by the high-efficiency coding apparatus of the embodiment of the present invention as described above. Optical disk, magneto-optical disk,
A disk-shaped recording medium such as a magnetic disk in which the above-mentioned encoded signal is recorded, a tape-shaped recording medium such as a magnetic tape in which the above-mentioned encoded signal is recorded, or a semiconductor in which the encoded signal is stored Examples include a memory and an IC card. In addition, examples of the transmission medium include an electric wire, an optical cable, and a radio wave.

The present invention is not limited only to this embodiment. For example, the above-mentioned recording / reproducing medium and the other recording / reproducing medium need not be integrated, and a data transfer between them is not necessary. It is also possible to connect with a line or the like. Further, for example, the present invention can be applied not only to an audio PCM signal but also to a signal processing device for a digital voice (speech) signal, a digital video signal, and the like. Further, the configuration may be such that the above-described minimum audible curve synthesis processing is not performed. In this case, the minimum audible curve generating circuit 532 and the synthesizing circuit 527 in FIG. 9 become unnecessary, and the output from the subtracter 526 is immediately transmitted to the subtractor 528.

Also, there are various bit allocation methods,
In the simplest case, it is possible to use fixed bit allocation, simple bit allocation based on the energy of each band of a signal, or bit allocation combining a fixed portion and a variable portion.

Further, in the memory 816 in FIG. 3, it is possible to reduce the number of quantization blocks held.
By reducing the number of blocks to be held, the capacity of the memory 816, the amount of calculation in the bit fraction adjustment circuit 814, and the like can be reduced, and the hardware scale can be reduced.

The primary bit distribution circuit 815 in FIG.
In the case where the output is a bit rate lower than a predetermined value of, for example, 128 kbps, the configuration may be such that the fraction adjustment of bits is not performed. In this case, the memory 816 and the bit fraction adjusting circuit 814 in FIG. 3 become unnecessary, and the output of the primary bit allocation circuit 815 immediately outputs the bit allocation output terminal 807.
Will be transmitted to

The primary bit distribution circuit 815 in FIG.
Is not limited to rounding, but it is also possible to perform rounding processing by changing the threshold value, for example, by rounding down 3 or less and rounding up 4 or more. For example, a method of simply performing round-up or round-down processing is also conceivable.

The embodiment of the present invention can have various modifications as described above.

[0090]

As is clear from the above description, the following effects can be obtained in the present invention. That is, when converting the bit information calculated as a decimal value for each two-dimensional block (quantized block) into an integer, the bit amount is determined to be equal to the bit amount determined by the predetermined bit rate in consideration of information on a rounding process performed. Adjust the bits so that
By determining the bit allocation amount for each two-dimensional block, it is possible to perform more adaptive bit allocation to the input signal, and it is possible to obtain good sound quality in terms of audibility at the same bit rate. Become. At a low bit rate, sound quality degradation can be prevented.

[Brief description of the drawings]

FIG. 1 is a block circuit diagram illustrating a configuration example of a high-efficiency encoding device according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a specific example of an orthogonal transform block size in a frequency and time domain in the apparatus according to the embodiment.

FIG. 3 is a block circuit diagram illustrating a configuration example of a bit allocation function according to an embodiment of the present invention.

FIG. 4 is a diagram showing bit information of each quantization block having a decimal part and the number of bits obtained by converting the bit information into an integer.

FIG. 5 is a diagram illustrating an example of bit fraction adjustment when a number of bits exceeding a predetermined bit rate is used.

FIG. 6 is a diagram illustrating an example of bit fraction adjustment using rounding information when converting bit information having a decimal part into an integer when the number of bits exceeding a predetermined bit rate is used.

FIG. 7 is a diagram illustrating an example of bit fraction adjustment when a bit number lower than a predetermined bit rate is used.

FIG. 8 is a diagram illustrating an example of bit fraction adjustment using rounding information when bit information having a decimal part is converted to an integer when a bit number lower than a predetermined bit rate is used.

FIG. 9 is a block circuit diagram showing a configuration example of an auditory masking threshold calculation function according to the embodiment of the present invention.

FIG. 10 is a diagram illustrating masking by each critical band signal.

FIG. 11 is a diagram illustrating a masking threshold by each critical band signal.

FIG. 12 is a diagram showing an information spectrum, a masking threshold, and a minimum audible limit.

FIG. 13 is a diagram showing a signal distribution and a permissible noise level-dependent bit allocation for an information signal having a flat signal spectrum.

FIG. 14 is a diagram showing signal level dependency and permissible noise level dependent bit allocation for an information signal having a high tonality of a signal spectrum.

FIG. 15 is a diagram showing a quantization noise level for an information signal having a flat signal spectrum.

FIG. 16 is a diagram illustrating a quantization noise level for an information signal having a high tonality.

FIG. 17 is a block circuit diagram illustrating a configuration example of a high-efficiency decoding device according to an embodiment of the present invention.

[Explanation of symbols]

 801 MDCT circuit output input terminal 802 Usable total bit generation circuit 803 Energy calculation circuit for each band 804 Energy-dependent bit allocation circuit 805 Bit allocation circuit depending on permissible noise level 806 Adder 807 Bit allocation output terminal for each band 808 Spectrum smoothness calculation circuit 809 Bit division ratio determination circuit 811 812 Multiplier 814 Bit fraction adjustment circuit 815 Primary bit distribution circuit 816 Memory

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-4-302532 (JP, A) JP-A-5-37396 (JP, A) JP-A-5-90973 (JP, A) JP-A-5-90973 206866 (JP, A) JP-A-6-6236 (JP, A) JP-A-7-135470 (JP, A) JP-A-7-143009 (JP, A) Japanese Translation of PCT Application No. 4-504192 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) H03M 7/30 G10L 19/00

Claims (7)

(57) [Claims] An input signal is decomposed into a plurality of frequency band components.
Frequency dividing step, and dividing the frequency band component with respect to time and frequency
According to the frequency band component for each of a plurality of two-dimensional blocks
Bit allocation information required for quantization is obtained as a decimal value, and
The above-mentioned bit allocation information is rounded and processed for each two-dimensional block.
Bit allocation calculation to calculate the bit allocation of integer values for each task
Output step and the bit allocation amount calculated for each of the two-dimensional blocks,
It is equal to the bit amount determined by the default bit rate
And determine the bit allocation amount for each two-dimensional block.
A bit allocation amount determining step, and the two-dimensional block according to the determined bit allocation amount.
Frequency band components that quantize the frequency band components
A quantization step, and in the bit allocation amount determining step, the two-dimensional block
The total bit allocation calculated for each clock is the default bit rate
If the bit amount is not equal to the
Based on the rounding information in the
And adjusting the bit allocation amount . In the bit allocation amount determining step, when the sum of the bit allocation amounts calculated for the two-dimensional blocks is larger than the bit amount determined by a predetermined bit rate, the bit allocation amount calculating step The number of bits corresponding to the rounded-up two-dimensional block has priority
2. The high-efficiency encoding method according to claim 1, wherein the number is reduced. In the bit allocation amount determining step, when the sum of the bit allocation amounts calculated for the two-dimensional blocks is smaller than the bit amount determined by a predetermined bit rate, the bit allocation amount calculating step The number of bits corresponding to the truncated two-dimensional block has priority
2. The high efficiency coding method according to claim 1, wherein the number is increased. 4. The method according to claim 1, wherein the information of the rounding process includes the bit
2. The method according to claim 1, wherein information on round-up and / or round-down processing in the allocation amount calculation step is used.
The high-efficiency encoding method according to the above. 5. The information of the rounding process, wherein the bit
2. The high-efficiency encoding method according to claim 1, wherein a numerical value of a digit after the decimal point of the bit allocation information obtained in the allocation amount calculating step is used . 6. An input signal is decomposed into a plurality of frequency band components.
Frequency dividing means, and the frequency band component is divided with respect to time and frequency.
According to the frequency band component for each of a plurality of two-dimensional blocks
Bit allocation information required for quantization is obtained as a decimal value, and
The above-mentioned bit allocation information is rounded and processed for each two-dimensional block.
Bit allocation calculation to calculate the bit allocation of integer values for each task
Output means, and the bit allocation amount calculated for each of the two-dimensional blocks,
It is equal to the bit amount determined by the default bit rate
And determine the bit allocation amount for each two-dimensional block.
Means for determining the bit allocation amount, and the two-dimensional block according to the determined bit allocation amount.
Frequency band components that quantize the frequency band components
And a quantizing means, wherein the bit allocation amount determining means is provided for each of the two-dimensional blocks.
The sum of the calculated bit allocation amounts depends on the default bit rate
If the bit amount is not equal to
Based on the information of the rounding process in the
A high-efficiency coding apparatus for adjusting the amount of data allocation . 7. The method according to claim 1 , wherein said compression coding is performed by a high efficiency coding method.
A transmission medium for transmitting an input signal, wherein the input signal is decomposed into a plurality of frequency band components, and the frequency band components are divided with respect to time and frequency.
According to the frequency band component for each of a plurality of two-dimensional blocks
Bit allocation information required for quantization is obtained as a decimal value, and
The above-mentioned bit allocation information is rounded and processed for each two-dimensional block.
The bit allocation amount of the integer value for each block is calculated, and the bit allocation amount calculated for each of the two-dimensional blocks is calculated by:
It is equal to the bit amount determined by the default bit rate
Bit allocation for each two-dimensional block by adjusting as
The two-dimensional block is determined in accordance with the determined bit allocation amount.
When quantizing the frequency band component for each clock,
The sum of the bit allocation amount calculated for each dimension block is the default
Is not equal to the bit rate determined by the
Is the information of the rounding process performed when calculating the bit allocation amount.
Is adjusted based on
Efficient coding method for determining block bit allocation
A transmission medium for transmitting a more compression-coded signal .