JP2001306095A - Device and method for audio encoding - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten the convergence time up to the determination of optimum 1st and 2nd scaling coefficients which satisfy a desired code quantity. SOLUTION: A scaling coefficient initial value control part 61 sets the initial value of the 1st scaling coefficient and the initial value of the 2nd scaling coefficient. A scaling coefficient update control part 62 updates the 2nd scaling coefficient in a direction, where the code quantity increases, evaluates a quantized noise quantity and the code quantity to find the optimum value of the 2nd scaling coefficient, and updates the 1st scaling coefficient in the increasing direction of the code quantity so that the difference between the code quantity and a target code quantity, when the 2nd scaling coefficient has an optimum value is put into a specific range, to find the optimum value of the 1st scaling coefficient.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio encoding apparatus and an audio encoding method for efficiently encoding an audio signal, and more particularly to an audio encoding method after orthogonally transforming an audio signal and compressing and encoding the audio signal by variable length encoding. Code amount,
The present invention relates to an audio encoding device and an audio encoding method for controlling convergence to a desired target code amount at high speed.

[0002]

2. Description of the Related Art In an audio encoding method, an audio signal is divided into a plurality of time blocks, and audio data is orthogonally transformed for each block to obtain a transform coefficient, which is scaled by an appropriate scaling coefficient (normalization). There is a method of quantizing the scaled transform coefficient and encoding the quantized value by variable length encoding using Huffman code or the like.

[0003] Such encoding methods include, for example, ISO
/ MPEG standardized as IEC 13818-7
-2 Advanced Audio Coding
There is an (AAC) coding scheme. In the coding amount control method in this encoding, in a frequency section in which a transform coefficient is divided into a plurality of blocks, a second scaling coefficient different from the first scaling coefficient common to all sections for the transform coefficient and different from each other in the audio section is used. This is to adaptively change the distribution of the quantization noise in consideration of the characteristics, and to bring the code amount after the variable length coding by the Huffman code at that time closer to a target code amount that is separately determined.

Here, the code amount is a bit amount required for encoding, and the target code amount is a value set for each block, for example, a predetermined encoded bit number. It is calculated from the rate, and may be fixed or variable for each block.

[0005] FIG. 8 is a diagram showing how the transform coefficient is scaled for each of a plurality of frequency band divisions. The input audio signal is transformed by orthogonal transform, and a transform coefficient for each frequency is obtained. Each transform coefficient is divided into a plurality of frequency band sections (here, seven sections from sb0 to sb6 are shown), and a scaling process is performed using the scaling coefficients. Here, the first scaling coefficient is a scaling coefficient common to all frequency band divisions, and the second scaling coefficient is a scaling coefficient different for each frequency band division.

FIG. 9 is a block diagram showing the configuration of a conventional audio encoding device specified in ISO / IEC 13818-7. An audio signal is input to the orthogonal transform unit 1 via an input terminal 101. This audio signal is usually time-series data, which is sampled at a specific sampling frequency and further quantized with a specific quantization bit. The orthogonal transform unit 1 performs a modified discrete cosine transform (MDCT: Modified Discr).
The time-series data is converted into frequency-series data by orthogonal transformation such as ET Cosine Transform. The data sequence after the orthogonal transform is called a transform coefficient. The converted conversion coefficient represents a frequency axis component, that is, spectrum data.

The audio signal from the input terminal 101 is
It is also input to the auditory characteristic analyzer 2. The auditory characteristic analyzer 2 performs an analysis modeling human auditory characteristics. One of the hearing characteristics is a characteristic called a masking effect, which is a phenomenon in which a certain sound makes other sounds inaudible. Using this characteristic, an index for adaptively controlling the amount of quantization noise generated is calculated so that the human ear does not perceive the quantization noise generated in the encoding process. This index is called a masking threshold, an allowable noise amount, or the like.

[0008] The auditory characteristic analyzer 2 also examines the characteristics of the input signal, and determines the block size of the orthogonal transform according to its steepness. This determines whether the resolution of the input signal in the orthogonal transform unit 1 is given priority over time resolution or frequency resolution. That is, as a result of analyzing the input signal, if the temporal change of the input signal is gradual, a longer block size is used to give priority to the frequency resolution, and if the temporal change of the input signal is sharp, In order to give priority to the resolution, a block size that is short in time is used.

The transform coefficient output from the orthogonal transform unit 1 is
It is divided into a plurality of frequency band sections, normalized by the scaling section 3, and adaptively quantized in the quantization section 4 for each of the frequency band sections. As the frequency band division, a division in which human auditory characteristics are considered is often used. In other words, a critical band division or the like having a wider bandwidth as the frequency becomes higher. Further, in the quantization for each frequency band section, normalization is performed in advance by the scaling unit 3 using a coefficient obtained for each frequency band section in order to increase the quantization efficiency.

[0010] The transform coefficients normalized for each frequency band section are quantized by the quantization unit 4. MPEG-2 AAC
In the method, the following equation (1) is used for normalization and quantization of transform coefficients. xq = int ((abs (coeff) * 2 ＾ ((1/4) * (scf-gscf))) ＾ 3/4 + α) (1) where xq is a quantization value, coeff is a transform coefficient, and gs
cf is a first scaling coefficient used for all bands when performing normalization, scf is a second scaling coefficient used for each frequency band division when performing normalization, and α is a correction value for quantization. , Scf and gscf are 0 or natural numbers. Note that int is a function that converts a decimal value to an integer value by truncation, and abs indicates a function that converts the value to an absolute value.

Further, the initial value scf_init of the second scaling coefficient scf and the initial value gscf_init of the first scaling coefficient gscf in the equation (1) are determined by the following equations (2) and (3). scf_init = 0 (2) gscf_init = (16/3) * log2 ((max_coeff ＾ (3/4)) / MAX_QUANT) (3) where max_coeff is a conversion coefficient that takes a maximum value among all conversion coefficients. MAX_QUANT is a constant indicating the maximum value range of the quantization value. Further, there is a restriction that the first scaling coefficient gscf must not be smaller than the initial value gscf_init of the first scaling coefficient depending on the condition of the range of the quantization value.

The quantized value obtained by the quantizing section 4 is variable-length coded by a variable-length coding section 5 using an optimum codebook for each band. The optimal codebook is a codebook in which the code amount after encoding is minimized.In order to obtain an accurate code amount for each codebook, encoding is performed for each selectable codebook. It is necessary to determine the code amount. Further, if the scaling coefficient in the equation (1) is updated, it is necessary to select a new optimal codebook again and recalculate the code amount at that time. In the MPEG-2 AAC system,
A Huffman codebook using Huffman codes is used.

The coding control unit 6 controls the scaling unit 3, the quantization unit 4, and the variable-length coding unit 5 in an integrated manner. The encoding control unit 6 specifies a second scaling coefficient scf for normalizing the scaling unit 3 based on the permissible noise amount for each frequency band segment obtained by the auditory characteristic analysis unit 2,
Also, the first scaling coefficient gscf for quantization by the quantization unit 4 is specified based on the magnitude of the code amount calculated by the variable length coding unit 5.

The first scaling coefficient gscf and the second scaling coefficient
By adaptively updating and controlling the two scaling coefficients of the scaling coefficient scf, the code amount after variable-length coding of the quantization value of the conversion coefficient is within the range of the target code amount, and the conversion coefficient is quantized. The trade-off relationship is adjusted so that the quantization noise generated at the time of performing the distribution has an optimal distribution perceptually. In general, the first scaling factor gscf
And the second scaling coefficient scf increases, the quantization value increases, the code amount increases, and the quantization noise decreases.

FIG. 10 is a flowchart showing the procedure of the process of the encoding control unit 6. In step ST101, an initial value gscf_init of the first scaling coefficient and an initial value scf_init of the second scaling coefficient are set by the above equations (3) and (2).
102, the first set in step ST103.
Using the initial value gscf_init of the scaling coefficient and the initial value scf_init of the second scaling coefficient,
Scaling and quantization are performed according to equation (1) to obtain a quantization value.

In step ST104, the obtained quantized value is subjected to variable length coding using an optimal codebook for each band, and in step ST105, the code amount at the time of performing variable length coding is calculated. An evaluation judgment with the target code amount is performed. If the code amount is larger than the target code amount, the judgment is NO, and if the code amount is smaller than the target code amount, the judgment is GO. In the case of a negative determination, step ST106
In, the first scaling coefficient is updated so that the code amount is smaller than the target code amount, and the process returns to step ST102.

If the determination is affirmative in step ST105, an evaluation determination regarding the amount of quantization noise is performed in step ST107, and if the distribution of quantization noise does not satisfy the acoustically optimal distribution condition, The judgment is made, and if the optimum distribution condition is satisfied, the judgment is a match. If a negative determination is made, in step ST108, the second scaling coefficient is updated so as to change the generation distribution of the quantization noise, and the process returns to step ST102.

FIG. 11 is a diagram showing how the quantization unit 4 performs quantization under the above initial value conditions. max_coeff
And the initial value gscf_init of the first scaling coefficient obtained based on MAX_QUANT and
The quantization value of ax_coeff is equivalent to MAX_QUANT. Under such conditions, when the quantized value is subjected to variable-length encoding, the quantized value takes a large value as a whole, and as a result, the code amount increases and exceeds the target code amount. Increase. Also, the code amount greatly varies depending on the value of each transform coefficient, that is, depending on the characteristics of the input signal.

Finally, the first scaling coefficient gscf and the second scaling coefficient scf are controlled so that quantization noise generated within the range of the target code amount is hardly perceived.
A series of processes of scaling, quantization, and variable-length coding are performed using the two scaling coefficients described above, and the code obtained after the processing is multiplexed by the multiplexing unit 7 in accordance with a format determined together with auxiliary information such as a scaling coefficient. Be transformed into

As described above, the conventional audio encoding apparatus orthogonally transforms audio data for each block,
The conversion coefficient is converted to a first scaling coefficient gscf and a second scaling coefficient gscf.
If the code amount obtained by scaling with the scaling coefficient scf, quantizing the transformed conversion coefficient, and further performing variable-length encoding does not become less than the desired target code amount, update the above two scaling coefficients and retry. That is, a so-called feedback method is adopted. In such a simple feedback method, the target code amount is determined by what values are selected for the scaling coefficient of the first trial, that is, the initial value gscf_init of the first scaling coefficient and the initial value scf_init of the second scaling coefficient. , Ie, the number of feedbacks is affected.

Also, depending on the method of updating the first scaling coefficient gscf and the second scaling coefficient scf, the code amount repeatedly over / unders the target code amount, so that the convergence time increases and the number of feedbacks increases. It will be.

[0022]

Since the conventional audio encoding apparatus is configured as described above, the optimum first audio encoding apparatus is used.
If the initial value gscf_init of the scaling coefficient and the initial value scf_init of the second scaling coefficient cannot be set, the convergence time until the optimal first scaling coefficient scf and second scaling coefficient scf for satisfying the desired code amount are obtained. However, there is a problem that encoding becomes impossible in real time. In addition, there is a problem that a high-performance arithmetic processing device is required to perform control at high speed. Such a problem is particularly remarkable when the target code amount is small, that is, at the time of low bit rate coding.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problem, and converges until the optimum first scaling coefficient scf and second scaling coefficient scf for satisfying a desired code amount by feedback are determined. An object of the present invention is to provide an audio encoding device and an audio encoding method in which time is shortened and code amount control converges at high speed.

[0024]

An audio encoding apparatus according to the present invention analyzes an input audio signal,
A hearing characteristic analysis unit that determines the block size of the orthogonal transformation and calculates an allowable noise amount for adaptively controlling the quantization noise amount, and performs an orthogonal transformation on the input audio signal, and the hearing characteristic analysis unit With the determined block size,
An orthogonal transform unit that outputs a transform coefficient that is a data sequence of a frequency sequence, and a frequency band division obtained by dividing the transform coefficient output from the orthogonal transform unit into a plurality of blocks. An initial value of a first scaling coefficient to be used, a scaling coefficient initial value control section for setting an initial value of a second scaling coefficient for normalization for each frequency band section, and a conversion coefficient output from the orthogonal transform section A scaling unit that performs normalization using the first scaling coefficient and the second scaling coefficient,
A quantization unit that quantizes the transform coefficient normalized by the scaling unit for each frequency band division; and a variable length coding that performs variable length coding of the transform coefficient quantized by the quantization unit for each frequency band division. Unit and the quantization noise amount after quantization by the quantization unit is evaluated based on the permissible noise amount calculated by the auditory characteristic analysis unit, and the second scaling coefficient exceeds the given target code amount. By updating the code amount in a direction in which the code amount increases, the optimum value of the second scaling coefficient is obtained, and the difference between the code amount when the second scaling coefficient is the optimum value and the target code amount is a predetermined value. And updating the first scaling coefficient in a direction in which the code amount increases so as to fall within the range, thereby providing a scaling coefficient update control unit for obtaining an optimum value of the first scaling coefficient. Those were example.

An audio encoding device according to the present invention comprises:
The scaling coefficient update control unit calculates the quantization noise amount after quantization by the quantization unit, evaluates the calculated quantization noise amount based on the permissible noise amount calculated by the auditory characteristic analysis unit, and calculates the quantization noise amount. Based on the evaluation result, the second scaling coefficient is provisionally updated in the increasing direction, and the code amount when the second scaling coefficient is provisionally updated is evaluated based on the target code amount. The update of the tentatively updated second scaling coefficient is permitted, and if the code amount is equal to or more than the target code amount, the update of the tentatively updated second scaling coefficient is stopped in the increasing direction to thereby optimize the second scaling coefficient. And updating the first scaling coefficient in a decreasing direction so that the difference between the code amount when the second scaling coefficient is the optimum value and the target code amount falls within a predetermined range. Ri is intended to obtain the optimum value of the first scaling factor.

An audio encoding device according to the present invention comprises:
When the quantization noise amount is evaluated based on the allowable noise amount, the second
When the update of the scaling coefficient is not necessary, the difference between the code amount at that time and the target code amount falls within a predetermined range.
The optimum value of the first scaling coefficient is obtained by updating the first scaling coefficient in the decreasing direction.

An audio encoding device according to the present invention comprises:
The scaling coefficient initial value control unit sets the initial value of the second scaling coefficient to “0” and sets the initial value of the first scaling coefficient to 1 as the quantization value for the transform coefficient having the maximum value.
The setting is as follows.

An audio encoding device according to the present invention comprises:
The scaling coefficient initial value control unit sets the initial value of the second scaling coefficient to a value as close as possible to 1 as long as the quantization value for the transform coefficient having the maximum value in each frequency band section is 1 or less. Then, the initial value of the first scaling coefficient is set so that the quantization value for the transform coefficient having the maximum value is 1 or less.

An audio encoding device according to the present invention comprises:
The scaling coefficient initial value control unit sets the initial value of the second scaling coefficient to “0”, or in a range where the quantization value for the transform coefficient having the maximum value in each frequency band section is 1 or less, infinitely. A value close to 1 is set, and the initial value of the first scaling coefficient is set to a value obtained by correcting the quantization coefficient for the transform coefficient having the maximum value to 1 or less and correcting the target code amount. To set.

An audio encoding method according to the present invention comprises:
Analyzing the input audio signal, and determining the block size of the orthogonal transform, the auditory characteristics analysis unit that calculates the allowable noise amount for adaptively controlling the quantization noise amount,
An orthogonal transform unit that orthogonally transforms an input audio signal, outputs a transform coefficient that is a data sequence of a frequency sequence in a block size determined by the auditory characteristic analysis unit, and a transform coefficient output from the orthogonal transform unit. A scaling unit that performs normalization using a first scaling coefficient common to all frequency bands and a second scaling coefficient for each frequency band division, and quantizes the transform coefficient normalized by the scaling unit for each frequency band division And a variable-length encoding unit that performs variable-length encoding of the transform coefficient quantized by the quantization unit for each frequency band division, and encodes the audio signal. A first step of setting an initial value of a scaling coefficient and an initial value of the second scaling coefficient; and setting the first scaling coefficient and the second scaling coefficient. A second step of performing normalization, quantization, and encoding by using the coefficient, and a quantization noise amount at the time of quantization in the second step, based on an allowable noise amount calculated by the auditory characteristic analysis unit. A third step of evaluating and confirming the necessity of updating the second scaling factor; and increasing the second scaling factor when the third step requires updating of the second scaling factor. A fourth step of temporarily updating the code amount in the direction and evaluating the code amount at the time of the temporary update based on the target code amount; and increasing the code amount if the code amount at the time of the temporary update in the fourth step is equal to or less than the target code amount. A fifth step of allowing the update of the second scaling coefficient provisionally updated in the direction and proceeding to the second step, and a code amount when the provisional update is performed in the fourth step is equal to or less than the target code amount. Ah For example, the updating of the second scaling coefficient provisionally updated in the increasing direction is stopped to thereby obtain an optimum value of the second scaling coefficient, and a code amount when the second scaling coefficient is the optimum value. And a seventh step of obtaining the optimum value of the first scaling coefficient by updating the first scaling coefficient in a decreasing direction so that the difference between the target code amounts falls within a predetermined range. is there.

[0031]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a configuration of an audio encoding device according to Embodiment 1 of the present invention. In the figure, the configuration other than the encoding control unit 60 is the same as that of the conventional audio encoding device shown in FIG. 9, and the description of the operation is omitted. 61 is a scaling coefficient initial value control unit provided in the coding control unit 60, and 62 is a coding control unit 6
This is a scaling coefficient update control unit provided in 0.

Next, the operation will be described. FIG. 2 is a flowchart showing the flow of the processing of the encoding control unit 60.
In step ST11, the scaling coefficient initial value control unit 61 uses the following equations (5) and (4) to set the first
Initial value gscf_init of scaling factor and second value
Set the initial value scf_init of the scaling factor. scf_init = 0 (4) gscf_init = (16/3) * log2 (max_coeff ＾ (3/4)) (5) The above equation (5) is an initial value gsc of the first scaling coefficient.
f_init is set so that the quantization value for the transform coefficient max_coeff that takes the maximum value among all the transform coefficients is 1 or less.

FIG. 3 is a diagram showing how the quantization unit 4 performs quantization under the above initial value conditions. Since the quantization value for max_coeff is defined to be 1 or less, the quantization values of all other transform coefficients are also 1 or less.
In such a case, if the quantized value is variable-length coded, the code amount is generally small, though it depends on the contents of the codebook used for variable-length coding.

Under the above initial value conditions, the transform coefficients output from the orthogonal transform unit 1 are subjected to steps ST12 to ST14.
In, a series of processes of scaling, quantization, and variable-length coding are performed. In step ST15, the scaling coefficient update control unit 62 of the encoding control unit 60 evaluates the amount of quantization noise for each frequency band section with reference to the quantization value and the code data output in the series of processing. , The optimum value of the second scaling coefficient is determined while evaluating the code amount for each frequency band section.

FIG. 4 is a flow chart showing a detailed processing procedure in step ST15 by the scaling coefficient update control section 62. In step ST21, the scaling coefficient update control unit 62 calculates the amount of quantization noise from the quantization result performed by the quantization unit 4, and in step ST22, evaluates the calculated amount of quantization noise to perform the second scaling. Check if the coefficient needs to be updated.

The evaluation of the quantization noise amount is to judge the degree of the masking effect from the relative relationship between the permissible noise amount obtained by the auditory characteristic analyzer 2 and the result of calculating the actual quantization noise amount. . For example, in the frequency band division,
Comparing the allowable noise amount and the magnitude of the quantization noise amount, if the quantization noise amount is smaller than the allowable noise amount, it is determined that the masking effect is sufficient, and the update of the second scaling coefficient is unnecessary, If it is determined that the update of the second scaling coefficient is unnecessary in all the frequency band sections, FIG.
The process proceeds to step ST17.

If it is determined in step ST22 that the second scaling coefficient needs to be updated, the process proceeds to step S22.
At T23, the scaling coefficient update control unit 62
A frequency band section in which the amount of quantization noise should be reduced is selected.
Here, one or a plurality of frequency band divisions may be selected at the same time. In step ST24, the scaling coefficient update control unit 62 tentatively updates the second scaling coefficient for the selected frequency band section so that the quantization noise amount decreases. For example, one is added to the current value of the second scaling coefficient scf,
The weighted value may be added in consideration of the relative relationship between the allowable noise amount and the quantization noise amount.

In step ST25, the scaling coefficient update control section 62 calculates the code amount when the second scaling coefficient scf is provisionally updated, and evaluates the code amount. As a result of the evaluation, if the code amount is smaller than the target code amount, in step ST26, the update of the second scaling coefficient is permitted, and in step ST16 of FIG.
The scaling coefficient update control unit 62 updates the second scaling coefficient scf and returns to step ST12.

If the code amount exceeds the target code amount in step ST25, the update of the second scaling coefficient scf is stopped in step ST27, and the process proceeds to step ST27.
At T28, the selected frequency band division is excluded from the selection of the frequency band division for which the second scaling coefficient scf needs to be updated thereafter.

In step ST29, it is confirmed whether or not all the bands in all the frequency band divisions have been excluded from the selection targets. If not all the bands have been excluded from the selection targets, the processes after step ST16 in FIG. 2 are performed. Step ST29
Then, when all the frequency band divisions are finally excluded from the selection target, the second scaling coefficient scf is determined to be optimized, the update processing of the second scaling coefficient scf is completed, and the step ST17 in FIG. Move to the processing of. The code amount is evaluated in step ST25, and if the code amount exceeds the target, the update of the second scaling coefficient scf is stopped in step ST27. It is always below the amount.

In the updating process of the second scaling coefficient scf, the updating of the first scaling coefficient gscf common to all frequency band divisions is not performed.
The recalculation of the quantization noise amount and the code amount may be performed only for the corresponding frequency band section.

In step ST22, it is determined that the update of the second scaling coefficient scf is unnecessary, and the optimization of the second scaling coefficient scf is completed. In step ST17 of FIG.
At the point in time, the code amount may be significantly lower than the target code amount. For example, if the input signal is a sine wave signal, it may be sufficient to reduce the quantization noise amount to less than the permissible noise amount only in the frequency band section containing the sine wave signal. The amount is much smaller than when the input signal contains various frequency components. Even in such a case, the scaling coefficient update control unit 62 performs update control on the first scaling coefficient gscf as the next step.

In step ST17 of FIG. 2, the scaling coefficient update control section 62 sets the second scaling coefficient s
The code amount at the end of the cf update process is compared with the target code amount, and it is checked whether the difference is within a predetermined range to confirm whether the first scaling coefficient gscf is optimal. If the difference is not within the predetermined range, in step ST18, the coding control unit 60 updates the first scaling coefficient, and repeats the processing in step ST12 and thereafter. This update is performed by, for example, the current first scaling factor gsc.
One may be subtracted from the value of f, or a value weighted according to the difference between the code amount and the target code amount may be subtracted.

In step ST17, if the difference between the code amount and the target code amount is within a predetermined range, it is determined that the first scaling coefficient gscf has been optimized, and the process ends. The predetermined range is a value set based on the amount of change in the code amount that changes with the update of the first scaling coefficient gscf, for example, and may be different for each frame.

In some cases, as described below, the first scaling coefficient gscf may be updated so that the code amount does not change. It is clear that the quantization value xq is invariable if the relationship of scf-gscf = constant (6) is maintained in the expression (1) which is a calculation expression of quantization. That is, the quantization noise amount and the code amount do not change. Based on this theory, in updating the first scaling coefficient gscf, if the update amount is β, the gscf is decreased by β, and at the same time, the second scaling coefficients scf of all the frequency band sections are also decreased by β. The update amount β at this time is the minimum update amount β that can finely control the quantization noise amount and the code amount.
= 1, but may be a value of β> 1 which can be changed greatly at a time to reduce the processing amount, or may be changed adaptively for each feedback loop.

In the feedback loop of the coding control unit 60, the updating of the second scaling coefficient scf and the first scaling coefficient gscf is repeated to obtain an optimum value.
If the code amount has converged, the process of the feedback loop has been completed. However, the code amount in the process of this feedback loop is controlled so as to be always lower than the target code amount. Therefore, in the updating processing of the second scaling coefficient scf and the first scaling coefficient gscf, it is only necessary to always perform the processing in the direction of increasing the code amount, and does not need the processing in the direction of decreasing the code amount. In other words, the second scaling coefficient scf and the first scaling coefficient gscf may be updated only in a direction in which the amount of quantization noise is improved. That is, the step of reducing the code amount can be omitted, and the convergence of the feedback loop does not need to be determined from both the over and under of the target code amount, making the determination easy.

In the above description of the code amount evaluation,
Although only the conversion coefficient that accounts for the majority of the code amount has been described in detail, it is added that, of course, the code amount evaluation also considers the change in the code amount necessary for auxiliary information such as a scaling coefficient. deep.

As described above, according to the first embodiment, the initial value gscf_init of the first scaling coefficient is set.
Is set so that the quantization value for the transform coefficient having the maximum value is 1 or less, and the initial value scf of the second scaling coefficient is set.
_Init is set to 0, and the first scaling coefficient gsc
In the updating process of f and the second scaling coefficient scf,
By always performing the updating process in the direction of increasing the code amount, the convergence time by feedback until the determination of the optimal first scaling coefficient gscf and the second scaling coefficient scf to satisfy the desired code amount can be shortened. The effect that code amount control can be converged at high speed is obtained.

Embodiment 2 The configuration of the audio encoding device according to this embodiment is the same as the configuration shown in FIG. 1 of the first embodiment, and the flow of processing of the encoding control unit 60 is as follows.
This is basically equivalent to the flowchart shown in FIG. 2 of the first embodiment.

Next, the operation will be described. In step ST11 of FIG. 2, the scaling coefficient initial value control unit 6
1 sets the initial value gscf_init of the first scaling coefficient and the initial value scf_init of the second scaling coefficient by the above equation (5) and the following equation (7).
scf_init = (16/3) * log2 ((max
_Coeff / sb_max_coeff) ＾ (3/4)) (7) sb_max_coeff in the above equation (7) is a transform coefficient that takes the maximum value for each frequency band section. This is shown in FIG.

FIG. 6 shows the state of quantization under the above initial value condition. As a result of setting the initial value scf_init of the second scaling coefficient by the above equation (7), the transform coefficient sb_max_coe having the maximum value in each frequency band section is set.
The quantization value for ff is within the range of 1 or less,
Take a value close to. That is, when considering the initial value scf_init of the second scaling coefficient for each frequency band section, the transform coefficient max that takes the maximum value in all frequency bands
_Coeff and sb_max in each frequency band division
By taking into account the relative ratio relationship of _coeff, normalization in each frequency band section can be performed efficiently. In other words, the quantization according to the above equation (1) corresponds to improving the quantization accuracy in advance as compared with the first embodiment.

When scaling, quantization, and variable length coding are performed under the above initial value conditions, the initial code amount is suppressed to a small amount, and the equation (4) of the first embodiment and the equation (4) of the second embodiment are used. When the equation (7) is compared, the initial value scf_init of the second scaling coefficient calculated by the equation (7) becomes larger. Therefore, the update processing of the second scaling coefficient scf is executed in advance as compared with the first embodiment. In other words, the processing amount of the entire feedback loop can be reduced accordingly.

The processing after step ST12 in FIG. 2 and the processing in FIG. 4 are the same as those in the first embodiment.

As described above, according to the second embodiment, the initial value gscf_init of the first scaling coefficient is set.
Is set so that the quantization value for the transform coefficient having the maximum value is 1 or less, and the initial value scf of the second scaling coefficient is set.
_Init is set such that the quantization value for the transform coefficient having the maximum value in each frequency band section takes a value as close as possible to 1 within a range of 1 or less, and the first scaling coefficient gs
In the updating process of the cf and the second scaling coefficient scf, the updating process in the direction of always increasing the code amount is performed, so that the optimal first scaling coefficient gscf and the second scaling coefficient scf to satisfy the desired code amount.
The convergence time due to the feedback up to the determination can be shortened, and the code amount control can be converged at high speed.

Embodiment 3 The configuration of the audio encoding device according to this embodiment is the same as the configuration shown in FIG. 1 of the first embodiment, and the flow of processing of the encoding control unit 60 is as follows.
This is basically equivalent to the flowchart shown in FIG. 2 of the first embodiment.

Next, the operation will be described. In step ST11 of FIG. 2, the scaling coefficient initial value control unit 6
1 sets the initial value gscf_init of the first scaling coefficient by the following equation (8) according to the value of the target code amount. gscf_init = 16/3 * log2 (max_coeff ＾ (3/4))-rate * γ (8) where rate is a target code amount and γ is a constant. Note that the initial value scf_init of the second scaling coefficient
Is a value set by the equation (4) of the first embodiment or the equation (7) of the second embodiment.

The above equation (8) is obtained by correcting the equation (5) of the first embodiment with the term rate * γ, and the initial value gscf_init of the first scaling coefficient is obtained.
Is set so that the quantization value for the transform coefficient having the maximum value is 1 or less, and is corrected to a target code amount. The value of the initial value gscf_init of the first scaling coefficient in the above equation (8) is rate
Due to the term * γ, the value is smaller than that in the case of the expression (5) in the first embodiment. Focusing on the quantization by the expression (1) in the case where the initial value gscf_init of the first scaling coefficient is used, if the value of the initial value gscf_init of the first scaling coefficient is small, the quantization in the entire frequency band division is performed. As the accuracy increases, the quantization noise decreases. In addition, a large quantization value requires a large amount of code.

FIG. 7 shows the first equation according to the equation (5) of the first embodiment.
It is a figure which shows the difference of code amount transition when the initial value gscf_init of a scaling coefficient is set, and when the initial value gscf_init of a 1st scaling coefficient is set by the said (8) formula. Scaling coefficient update controller 6
2 for the code amount after the first loop processing, the initial value gscf of the first scaling coefficient by the above equation (5).
If the case where _init is set is m and the case where the initial value gscf_init of the first scaling coefficient is set by the above equation (8) is n, then the relationship between m and n is m <n.

Thereafter, the code amount is set to the target code amount ra.
The feedback loop is repeated until te converges.
Regarding the number of repetitions of the feedback loop when the code amount converges to the target code amount rate, let M be the case where the initial value gscf_init of the first scaling coefficient is set by the above equation (5), and make the first scaling by the above equation (8). When the initial value gscf_init of the coefficient is set as N, the relationship between M and N at this time is M> N.
This means that the process in the feedback loop performed by the encoding control unit 60 is
This is equivalent to executing in advance.

When scaling, quantization, and variable length coding are performed under the above initial value condition, the first scaling coefficient is updated while the initial code amount is kept small and compared with the first embodiment. Is executed in advance, and the processing amount of the entire feedback loop can be suppressed by that amount.

The processing after step ST12 in FIG. 2 and the processing in FIG. 4 are the same as those in the first embodiment.

As described above, according to the third embodiment, the initial value gscf_init of the first scaling coefficient is set.
Is set so that the quantization value for the transform coefficient having the maximum value is 1 or less, and is set to a value corrected by the target code amount, and the initial value scf_i of the second scaling coefficient is set.
nit is set to 0, or the quantization value for the transform coefficient having the maximum value in each frequency band section is set to take a value as close as possible to 1 within a range of 1 or less, and the first scaling coefficient gscf and second scaling coefficient scf_
In the updating process of init, by always performing the updating process in the direction of increasing the code amount, the feedback by the feedback up to the determination of the optimal first scaling coefficient gscf and the second scaling coefficient scf for satisfying the desired code amount is performed. The effect is obtained that the convergence time can be shortened and the code amount control can be converged at high speed.

[0063]

As described above, according to the present invention, the initial value of the first scaling coefficient used in common for all frequency bands at the time of normalization for the frequency band division obtained by dividing the transform coefficient into a plurality of blocks. And a scaling coefficient initial value control unit for setting an initial value of a second scaling coefficient for performing normalization for each frequency band section, and evaluating the quantization noise amount based on the allowable noise amount, and setting the second scaling coefficient , The code amount is updated so as not to exceed the given target code amount, so that the optimum value of the second scaling coefficient is obtained, and the code amount when the second scaling coefficient is the optimum value and the target code The first scaling coefficient is updated in a direction in which the code amount increases so that the difference in the amount falls within a predetermined range, thereby performing scaling for obtaining an optimum value of the first scaling coefficient. By providing the number update control unit, the convergence time due to feedback until the optimal first scaling coefficient and the second scaling coefficient for satisfying the desired code amount can be shortened, and the code amount control can be quickly converged. There is an effect that can be.

According to the present invention, the scaling coefficient update control unit evaluates the calculated quantization noise amount based on the allowable noise amount, and based on the quantization noise amount evaluation result, increases the second scaling coefficient in the increasing direction. The code amount when the provisional update is performed and the second scaling coefficient is provisionally updated is evaluated based on the target code amount. If the code amount is equal to or less than the target code amount, the provisional update of the provisionally updated second scaling coefficient in the increasing direction is permitted. If the code amount is equal to or more than the code amount, the update of the second scaling coefficient provisionally updated in the increasing direction is stopped to obtain the optimum value of the second scaling coefficient. By updating the first scaling coefficient in a decreasing direction so that the difference in code amount falls within a predetermined range, a desired value of the first scaling coefficient is obtained, Convergence time by optimal first scaling factor to satisfy the issue amount and feedback to the determination of the second scaling factor to be shortened, there is an effect that it is possible to converge the code amount control at high speed.

According to the present invention, when the quantization noise amount is evaluated based on the permissible noise amount, if it is not necessary to update the second scaling coefficient, the difference between the code amount at that time and the target code amount is determined. By updating the first scaling coefficient in the decreasing direction so as to fall within the range, the optimum value of the first scaling coefficient is obtained. 2 There is an effect that the convergence time due to feedback up to the determination of the scaling coefficient can be shortened, and the code amount control can be converged at high speed.

According to the present invention, the scaling coefficient initial value control unit sets the initial value of the second scaling coefficient to “0” and sets the initial value of the first scaling coefficient to the quantization coefficient for the transform coefficient having the maximum value. By setting the value to be 1 or less, it is possible to shorten the convergence time due to feedback until the optimal first scaling coefficient and second scaling coefficient for satisfying the desired code amount are determined, and the code amount control is performed at high speed. Has the effect of being able to converge.

According to the present invention, the scaling coefficient initial value control unit sets the initial value of the second scaling coefficient to a maximum value within a range where the quantization value for the transform coefficient having the maximum value in each frequency band section is 1 or less. And a value close to 1 is set, and the initial value of the first scaling coefficient is set so that the quantization value for the transform coefficient having the maximum value is 1 or less, thereby satisfying a desired code amount. The convergence time due to feedback until the optimal first scaling coefficient and second scaling coefficient are determined can be shortened, and the code amount control can be converged at high speed.

According to the present invention, the scaling coefficient initial value control unit sets the initial value of the second scaling coefficient to “0” or sets the quantization value for the transform coefficient having the maximum value in each frequency band section. Is in the range of 1 or less,
, And the initial value of the first scaling coefficient is set to 1 as the quantization value for the transform coefficient having the maximum value.
By setting the value to a value corrected by the target code amount, convergence by feedback until the determination of the optimal first and second scaling coefficients for satisfying the desired code amount is performed. There is an effect that time can be shortened and code amount control can be converged at high speed.

According to the present invention, the first step of setting the initial value of the first scaling coefficient and the initial value of the second scaling coefficient,
A second step of performing normalization, quantization, and encoding using the scaling coefficient, and evaluating a quantization noise amount at the time of quantization in the second step based on the allowable noise amount, and updating the second scaling coefficient. A third step of confirming the necessity, and, if there is a need to update the second scaling coefficient in the third step, the second scaling coefficient is provisionally updated in the increasing direction, and the code amount at the time of the provisional update is reduced. A fourth step of evaluating based on the target code amount, and if the code amount at the time of the provisional update in the fourth step is equal to or smaller than the target code amount, updating of the second scaling coefficient provisionally updated in the increasing direction is permitted. A fifth step for shifting to the second step, and a fourth step for
If the code amount at the time of the provisional update in the step is less than or equal to the target code amount, the update of the provisionally updated second scaling coefficient in the increasing direction is stopped, thereby obtaining an optimum value of the second scaling coefficient. And updating the first scaling coefficient in the decreasing direction so that the difference between the code amount when the second scaling coefficient is the optimum value and the target code amount falls within a predetermined range, thereby achieving the optimum value of the first scaling coefficient. By providing the seventh step of obtaining the value, the convergence time due to feedback until the optimal first scaling coefficient and the second scaling coefficient for satisfying the desired code amount can be shortened, and the code amount control can be performed. There is an effect that convergence can be achieved at high speed.

[Brief description of the drawings]

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

FIG. 2 is a flowchart showing a processing flow of an encoding control unit according to the first embodiment of the present invention.

FIG. 3 is a diagram showing a state of quantization performed by a quantization unit according to the first embodiment of the present invention.

FIG. 4 is a flowchart illustrating a detailed processing procedure of a scaling coefficient update control unit according to the first embodiment of the present invention.

FIG. 5 is a diagram showing transform coefficients for each frequency band section according to Embodiment 2 of the present invention.

FIG. 6 is a diagram illustrating a state of quantization performed by a quantization unit according to a second embodiment of the present invention.

FIG. 7 is a diagram illustrating a difference in code amount transition due to a difference in an initial value of a first scaling coefficient according to the third embodiment of the present invention.

FIG. 8 is a diagram illustrating a state in which a transform coefficient is scaled for each of a plurality of frequency band divisions.

FIG. 9 is a block diagram illustrating a configuration of a conventional audio encoding device.

FIG. 10 is a flowchart showing a procedure of processing of a conventional encoding control unit.

FIG. 11 is a diagram illustrating a state of quantization performed by a conventional quantization unit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Orthogonal transformation part, 2 auditory-characteristic analysis part, 3 scaling part, 4 quantization part, 5 variable-length coding part, 7 multiplexing part, 60 coding control part, 61 scaling coefficient initial value control part, 62 scaling coefficient update control Part, 101
Input terminal, 102 output terminal.

Claims (7)

[Claims] 1. An auditory characteristic analyzer for analyzing an input audio signal, determining an orthogonal transform block size, and calculating an allowable noise amount for adaptively controlling a quantization noise amount. Orthogonal transformation of the audio signal, and an orthogonal transformation unit that outputs a transformation coefficient, which is a data sequence of a frequency sequence, in a block size determined by the auditory characteristic analysis unit, and a plurality of transformation coefficients output from the orthogonal transformation unit. For the frequency band divisions divided into blocks, the initial value of the first scaling coefficient used in common for all frequency bands at the time of normalization and the initial value of the second scaling coefficient for performing normalization for each frequency band division A scaling coefficient initial value control unit to be set, and a conversion coefficient output from the orthogonal transformation unit, using the first scaling coefficient and the second scaling coefficient. A quantization unit that quantizes the transform coefficient normalized by the scaling unit for each frequency band division; and a quantization unit that quantizes the transform coefficient quantized by the quantization unit for each frequency band division. A variable-length encoding unit for performing variable-length encoding, and a quantization noise amount after quantization by the quantization unit is evaluated based on an allowable noise amount calculated by the auditory characteristic analysis unit, and the second scaling coefficient is calculated. By updating the code amount in a direction in which the code amount increases so as not to exceed the given target code amount, the optimum value of the second scaling coefficient is obtained, and the code amount when the second scaling coefficient is the optimum value, In order that the difference between the target code amounts falls within a predetermined range,
An audio encoding device, comprising: a scaling coefficient update control unit that obtains an optimum value of the first scaling coefficient by updating the first scaling coefficient in a direction in which a code amount increases. 2. A scaling coefficient update control unit calculates a quantization noise amount after quantization by the quantization unit, and evaluates the calculated quantization noise amount based on an allowable noise amount calculated by the auditory characteristic analysis unit. Based on the evaluation result of the quantization noise amount, the second scaling coefficient is provisionally updated in the increasing direction, and the code amount when the second scaling coefficient is provisionally updated is evaluated based on the target code amount. If there is, the update of the second scaling coefficient provisionally updated in the increasing direction is permitted, and if the code amount is equal to or larger than the target code amount, the update of the second scaling coefficient provisionally updated in the increasing direction is stopped, thereby updating the second scaling coefficient. Calculating the optimum value of the scaling coefficient, and the code amount when the second scaling coefficient is the optimum value;
2. The audio code according to claim 1, wherein an optimum value of the first scaling coefficient is obtained by updating the first scaling coefficient in a decreasing direction so that a difference between the target code amounts falls within a predetermined range. Device. 3. When the quantization noise amount is evaluated based on the permissible noise amount and the update of the second scaling coefficient is not necessary, the difference between the code amount at that time and the target code amount falls within a predetermined range. 3. The audio encoding apparatus according to claim 2, wherein the optimum value of the first scaling coefficient is obtained by updating the first scaling coefficient in a decreasing direction so as to be included. 4. A scaling coefficient initial value control section sets an initial value of a second scaling coefficient to “0”,
2. The audio encoding apparatus according to claim 1, wherein an initial value of the scaling coefficient is set such that a quantization value for a transform coefficient having a maximum value is 1 or less. 5. A scaling coefficient initial value control unit sets an initial value of the second scaling coefficient to a value close to 1 in a range where a quantization value for a transform coefficient having a maximum value in each frequency band section is 1 or less. 2. The audio encoding apparatus according to claim 1, wherein the audio encoding apparatus is set to take a value, and the initial value of the first scaling coefficient is set so that a quantization value for a transform coefficient having a maximum value is 1 or less. . 6. A scaling factor initial value control unit sets an initial value of a second scaling factor to “0”,
Alternatively, the quantization value for the transform coefficient having the maximum value in each frequency band section is set to be as close as possible to 1 within a range of 1 or less, and the initial value of the first scaling coefficient is set to the maximum value. 2. The audio encoding apparatus according to claim 1, wherein the quantization value for the transform coefficient is set to 1 or less, and the value is set to a value corrected by the target code amount. 7. An auditory characteristic analyzer for analyzing an input audio signal, determining an orthogonal transform block size, and calculating an allowable noise amount for adaptively controlling a quantization noise amount. An orthogonal transform unit that orthogonally transforms the converted audio signal and outputs a transform coefficient that is a data sequence of a frequency sequence in a block size determined by the auditory characteristic analysis unit; and a transform coefficient output from the orthogonal transform unit. A scaling unit that performs normalization using a first scaling coefficient common to frequency bands and a second scaling coefficient for each frequency band division; and a quantum that quantizes the transform coefficient normalized by the scaling unit for each frequency band division. An audio signal, comprising: a coding section; and a variable-length coding section that performs variable-length coding on the transform coefficient quantized by the quantization section for each frequency band section. A first step of setting an initial value of the first scaling factor and an initial value of the second scaling factor; and setting the first scaling factor and the second scaling factor. And the second step of performing normalization, quantization and encoding, and the quantization noise amount when quantizing in the second step
A third step of evaluating based on the permissible noise amount calculated by the auditory characteristic analyzer and confirming the necessity of updating the second scaling coefficient; and a necessity of updating the second scaling coefficient in the third step. A fourth step of tentatively updating the second scaling coefficient in an increasing direction and evaluating a code amount at the time of the tentative update based on a target code amount, and a tentative update in the fourth step. If the code amount is less than or equal to the target code amount, the fifth step of allowing the update of the second scaling coefficient provisionally updated in the increasing direction and proceeding to the second step, and the fourth step If the code amount at the time of the provisional update is equal to or less than the target code amount, the update of the second scaling coefficient provisionally updated in the increasing direction is stopped, and the second
A sixth step of obtaining an optimum value of the scaling coefficient; a code amount when the second scaling coefficient is an optimum value;
And updating the first scaling coefficient in a decreasing direction so that the difference between the target code amounts falls within a predetermined range, thereby obtaining an optimum value of the first scaling coefficient. Audio encoding method to use.