JP6492915B2 - Encoding apparatus, encoding method, and program - Google Patents

Description

  The present invention relates to an encoding device, an encoding method, and a program.

  As a coding method of an audio signal or a sound signal (hereinafter collectively referred to as “audio signal”), there is a method of reducing the amount of information using human auditory characteristics such as an Advanced Audio Cording (AAC) method. In this type of encoding method, the number of bits required for encoding without increasing perceivable noise by suppressing the quantization error that increases when an audio signal is quantized with a small number of bits to a predetermined masking threshold or less (no increase in perceivable noise) That is, the amount of information) is reduced.

  The ideal value of the masking threshold is the upper limit value of the quantization error amount that cannot be perceived by humans. Therefore, the masking threshold is calculated based on the psychoacoustic model. Hereinafter, an ideal masking threshold calculated based on the psychoacoustic model is referred to as an initial masking threshold.

  However, in coding under a low bit rate condition (for example, 64 kbps or less), since the number of usable bits is small, the quantization error cannot often be suppressed below the initial masking threshold. If the quantization error cannot be suppressed below the initial masking threshold, the initial masking threshold is corrected based on the bit rate condition. (For example, refer nonpatent literature 1.).

  In addition, as a method for efficiently using a limited amount of bits in encoding under a low bit rate condition, there is known a method for adaptively switching the allocation of the number of bits to each band in a fixed or variable manner according to the audio signal. (For example, see Patent Document 1).

JP 2000-151413 A

"3GPP TS 26.403 V9.0.0", [online], 3GPP, March 8, 2015 search, Internet <URL: http://www.arib.or.jp/IMT-2000/V900Jul11/5_Appendix/Rel9/ 26 / 26403-900.pdf>

  Masking threshold correction methods are roughly classified into a method that is performed under a condition that allows a band to be lost due to quantization, and a method that is performed under a condition that does not allow a band to be lost.

  When the masking threshold is corrected under a condition that allows the loss of bands, if the correction amount increases, a band including sound that can be perceived by humans may be lost due to quantization. If a band including a sound that can be perceived by quantization (encoding) is lost, when the encoded audio signal is reproduced (decoded), a person who listens to the reproduced sound is uncomfortable. For this reason, when the number of missing bands increases, the sound quality deteriorates.

  On the other hand, when the masking threshold value is corrected under a condition that does not allow the loss of the band, an upper limit value is set for the masking threshold value of each band. Therefore, when the correction amount reaches the upper limit and a band that cannot be further corrected is generated, the correction amount of that band cannot be increased (in other words, the number of allocated bits cannot be reduced), and the correction amount of other bands is increased. It becomes. Therefore, the masking threshold value for the band having a large difference between the initial masking threshold value and the upper limit value is excessively corrected, and the number of bits allocated for encoding the band is reduced. A band having a large difference between the initial masking threshold and the upper limit value is a band important for sound quality. That is, when the masking threshold is corrected under conditions that do not allow band loss, the number of bits allocated to a band important for sound quality decreases, leading to deterioration of sound quality.

  In one aspect, an object of the present invention is to suppress deterioration in sound quality when an audio signal is encoded using a masking threshold based on auditory characteristics.

  An encoding device according to one aspect of the present invention is an encoding device that converts an audio signal into a frequency spectrum, and performs quantization and encoding of the frequency spectrum. A threshold generation unit, a threshold correction unit, and a recorrection A control unit. The threshold generation unit generates an initial masking threshold for quantization based on the frequency spectrum. The threshold correction unit corrects the initial masking threshold based on a bit amount usable for quantization of the frequency spectrum. The re-correction control unit determines whether or not to allow omission for each band of the frequency spectrum when the approximate similarity between the masking threshold corrected by the threshold correction unit and the initial masking threshold is equal to or less than a reference value. Then, the threshold value correction unit corrects the initial masking threshold value again.

  According to the above-described aspect, it is possible to suppress deterioration in sound quality when an audio signal is encoded using a masking threshold based on auditory characteristics.

It is a block diagram which shows the structural example of the encoding apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart (the 1) which shows the encoding process in the encoding apparatus which concerns on 1st Embodiment. It is a flowchart (the 2) which shows the encoding process in the encoding apparatus which concerns on 1st Embodiment. It is a flowchart (the 3) which shows the encoding process in the encoding apparatus which concerns on 1st Embodiment. It is a flowchart which shows the content of the missing | missing tolerance band setting process in 1st Embodiment. It is a graph which shows an example of a frequency spectrum and an initial masking threshold value. It is a graph explaining the relationship between the masking threshold value obtained by the first correction and the initial masking threshold value. It is a graph explaining the relationship between the masking threshold value obtained by the second correction and the masking threshold value obtained by the first correction. It is a graph explaining the relationship between the masking threshold value obtained by the third correction and the masking threshold value obtained by the second correction. It is a schematic diagram which shows the hardware structural example of the computer used as an encoding apparatus. It is a flowchart which shows the modification of the correction process of the masking threshold value which concerns on 1st Embodiment. It is a flowchart which shows the correction process of the masking threshold value which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows the content of the missing | missing tolerance band setting process in 2nd Embodiment. It is a graph explaining the determination method of the ratio of the zone | band | zone which accept | permits a loss | missing. It is a flowchart which shows the content of the deletion | missing tolerance band setting process which concerns on the 3rd Embodiment of this invention. It is a graph explaining the determination method of the ratio of the zone | band | zone which accept | permits a loss | missing.

[First Embodiment]
FIG. 1 is a block diagram showing a configuration example of an encoding apparatus according to the first embodiment of the present invention.

  As shown in FIG. 1, the encoding apparatus 1 according to the present embodiment includes a block switching unit 10, an MDCT processing unit 12, a masking threshold generation unit 14, an auditory characteristic calculation unit 16, a masking threshold correction unit 18, and a recorrection control unit 20. , A quantization unit 22, an encoding unit 24, and a multiplexing unit 26.

  The block switching unit 10 switches the block length when performing Modified Discrete Cosine Transform (MDCT) processing on the input signal based on the characteristics of the input signal (audio signal). In AAC coding, the block length is switched to a long block (1024 points) or a short block (128 points).

  The MDCT processing unit 12 performs MDCT processing with a window length corresponding to the long block or the short block on the input signal, and converts the input signal into a frequency spectrum. In AAC encoding, if the block length is a long block, MDCT processing with a window length of 2048 is performed, and if the block length is short, MDCT processing with a window length of 256 is performed.

The masking threshold generation unit 14 performs auditory psychological analysis on the input signal, and generates an optimal masking threshold (initial masking threshold) sfbThr ₀ in the quantization of the frequency spectrum obtained from the input signal. The initial masking threshold sfbThr ₀ is generated for each scale factor band sfb (hereinafter also referred to as “band sfb”) in the frequency spectrum. Further, the masking threshold value generator 14 determines the band sfb to be encoded based on the power value (input power) mdct_pow (sfb) and the initial masking threshold value sfbThr ₀ (sfb) in each band sfb of the frequency spectrum. Further, after determining the encoding target band sfb, the masking threshold generation unit 14 determines whether or not quantization using the initial masking threshold is possible, in other words, whether or not the masking threshold needs to be corrected. .

  The auditory characteristic calculation unit 16 calculates the auditory characteristic necessary for correcting the masking threshold. The auditory characteristic calculation unit 16 of the present embodiment calculates a signal mask ratio (SMR) of each band sfb as the auditory characteristic.

  The masking threshold correction unit 18 corrects the masking threshold based on the auditory characteristics (signal to mask ratio) and the upper limit value of the masking threshold set for each band sfb. Note that the masking threshold value correction unit 18 in the present embodiment, when correction for a set of initial masking threshold values is performed for the first time, sets the upper limit value of the masking threshold values of all the bands sfb to be missing due to quantization. Set to a value that does not allow. As will be described later, when correction for a set of initial masking threshold values is performed for the second time or later, the masking threshold value correction unit 18 performs correction under a condition that allows some bands to be lost.

  The re-correction control unit 20 determines whether or not to use the corrected masking threshold value, and if not adopted, causes the masking threshold value correction unit 18 to correct the masking threshold value again. When the masking threshold is corrected again, the recorrection control unit 20 sets a band that allows missing due to quantization. The re-correction control unit 20 includes a acceptance / rejection determination unit 20a, a missing allowable band setting unit 20b, and a storage unit 20c.

The acceptance / rejection determination unit 20a determines acceptance / rejection of the corrected masking threshold. In the present embodiment, whether or not the masking threshold is adopted is determined based on the approximate similarity between the corrected masking threshold and the initial masking threshold. For the approximate similarity, a cross-correlation value between the corrected masking threshold sfbThr (sfb) and the initial masking threshold sfbThr ₀ (sfb) is used.

  The missing permissible band setting unit 20b sets a band that allows omission when the acceptance / rejection determination unit 20a determines that it is not adopted. In the present embodiment, the loss of the band with the lowest importance is permitted among the bands in which the upper limit value of the masking threshold is set to a value that does not allow the loss. For the importance of the band, a signal-to-mask ratio (SMR) is used, and a band having a higher importance as the signal-to-mask ratio is larger.

The storage unit 20c stores information necessary for recorrecting the masking threshold. In this embodiment, information including the initial masking threshold value sfbThr ₀ (sfb), the signal-to-mask ratio, the upper limit value of the masking threshold value, and the importance level for each band sfb to be encoded is stored.

  The quantization unit 22 quantizes the band to be encoded in the frequency spectrum using either the initial masking threshold or the corrected masking threshold.

  The encoding unit 24 encodes a value obtained by quantizing the frequency spectrum. In the case of encoding by the AAC method, the encoding unit 24 performs Huffman encoding on the value obtained by quantization.

  The multiplexing unit 26 multiplexes the encoded audio signal to generate an encoded stream.

Next, the encoding process in the encoding device 1 according to the present embodiment will be described.
FIG. 2A is a flowchart (part 1) illustrating an encoding process in the encoding device according to the first embodiment. FIG. 2B is a flowchart (part 2) illustrating the encoding process in the encoding device according to the first embodiment. FIG. 2C is a flowchart (part 3) illustrating the encoding process in the encoding device according to the first embodiment.

  The encoding apparatus 1 according to the present embodiment performs encoding processing as shown in FIGS. 2A to 2C on each encoding unit data such as a frame in an input signal (audio signal).

  As shown in FIG. 2A, the encoding device 1 first converts an input signal for one frame into a frequency spectrum, and calculates a power value mdct_pow (sfb) of each band sfb (step S10). The block switching unit 10 and the MDCT processing unit 12 perform the process in step S10.

  The block switching unit 10 selects and switches whether the block length of the MDCT processing is a long block or a short block. The block length is selected based on a known selection method, for example, an input signal power fluctuation ratio and a predicted gain fluctuation ratio.

  Further, the MDCT processing unit 12 performs an MDCT process with a window length corresponding to the block length selected by the block switching unit 10 and converts the input signal into a frequency spectrum. The MDCT processing unit 12 converts the input signal xin into the frequency spectrum mdct (k) by the following equation (1), for example.

  N in Formula (1) is the window length of MDCT processing. In the case of encoding by the AAC method, if the block length is a long block, the window length N is set to 2048, and if the block length is a short block, the MDCT process is performed with the window length N set to 256.

  Further, the MDCT processing unit 12 calculates the power value mdct_pow (sfb) of each band (scale factor band sfb) based on the obtained frequency spectrum. The power value mdct_pow (sfb) is calculated by the following equation (2), for example.

  Note that the conversion of the audio signal into the frequency spectrum is not limited to the conversion using Equation (1), and any one of known conversion methods may be used. Similarly, the power value mdct_pow (sfb) of each band sfb of the frequency spectrum is not limited to the equation (2), and may be calculated using any known calculation method.

Next, the encoding device 1 generates an initial masking threshold sbfThr ₀ (sfb) for quantizing the frequency spectrum (step S12). The process of step S12 is performed by the masking threshold value generator 14.

The masking threshold value generation unit 14 performs auditory psychological analysis on the input signal and obtains an initial masking threshold value sfbThr ₀ (sfb) for each band sfb. The initial masking threshold value sfbThr ₀ (sfb) is calculated using any known calculation method based on the minimum audible level, the masking effect, etc. in each band sfb.

In addition, when the masking threshold value generator 14 generates the initial masking threshold value sfbThr ₀ (sfb), the masking threshold value generator 14 next generates an encoding target based on the initial masking threshold value sfbThr ₀ (sfb) and the frequency spectrum power value mdct_pow (sfb). A band is determined (step S14). In the process of step S14, the masking threshold value generation unit 14 sets only the band of sfbThr ₀ (sfb) <mdct_pow (sfb) out of the entire band of the frequency spectrum as an encoding target.

  After determining the encoding target band, the masking threshold generation unit 14 calculates an initial PE value and a target PE value in order to determine whether or not to correct the masking threshold (step S16). In the present embodiment, it is determined whether or not to correct the masking threshold based on whether or not the initial PE value is larger than the target PE value (step S18).

Here, the PE value is a value of perceptual entropy, which is one of acoustic parameters, and represents difficulty in encoding. The initial PE value is a PE value calculated based on the power value mdct_pow (sfb) and the initial masking threshold sfbThr ₀ (sfb) in the band to be encoded. The target PE value is a PE value calculated based on the number of bits that can be used for encoding. The initial PE value and the target PE value are calculated using any known calculation method (for example, the calculation method described in Non-Patent Document 1).

  The perceptual entropy value is related to the number of bits necessary for quantization as described above, and when the initial PE value is larger than the target PE value, the bit amount used for quantization using the initial masking threshold is bit. It can be determined that the number of bits given according to the rate is exceeded. On the other hand, when the initial PE value is less than or equal to the target PE value, it can be determined that the bit amount used in quantization using the initial masking threshold falls within the number of bits given according to the bit rate. Therefore, based on the magnitude relationship between the initial PE value and the target PE value, it can be determined whether quantization using the initial masking threshold is possible, that is, whether the masking threshold needs to be corrected.

When the initial PE value is less than or equal to the target PE value (step S18; No), the masking threshold value generation unit 14 determines that the masking threshold value is not corrected, and passes the initial masking threshold value sfbThr ₀ (sfb) to the quantization unit 22. In this case, the encoding device 1 quantizes the frequency spectrum using the initial masking threshold sfbThr ₀ (sfb) as shown in FIG. 2C (step S32). The quantization unit 22 performs the quantization in step S32. The quantization unit 22 quantizes the frequency spectrum using any of the known quantization methods.

On the other hand, when the initial PE value is larger than the target PE value (step S18; Yes), the masking threshold value generation unit 14 determines that the masking threshold value is to be corrected, and the initial masking threshold value sfbThr ₀ (sfb) and the target PE value are determined as auditory characteristics. It passes to the calculation part 16 and the acceptance / rejection determination part 20a. In this case, the encoding apparatus 1 performs a masking threshold value correction process as in steps S20 to S30 illustrated in FIG. 2B.

  When correcting the masking threshold, the encoding apparatus 1 next calculates an auditory characteristic based on the frequency spectrum or the like (step S20). The process of step S20 is performed by the auditory characteristic calculation unit 16.

The auditory characteristic calculator 16 calculates a signal-to-mask ratio (SMR) in each band sfb, that is, a difference value between the power value mdct_pow (sfb) and the initial masking threshold sfbThr ₀ (sfb) in each band. When the signal-to-mask ratio is calculated, the auditory characteristic calculation unit 16 passes the initial masking threshold value sfbThr ₀ (sfb), the target PE value, and the signal-to-mask ratio to the masking threshold value correction unit 18 and recorrects the signal-to-mask ratio. The data is stored in the storage unit 20c of the control unit 20.

  Next, the encoding device 1 performs a loss prevention process and a masking threshold correction process. The masking threshold correction unit 18 performs the omission prevention processing and the masking threshold correction processing.

  The loss prevention process is a process for preventing a band from being lost when a frequency spectrum is quantized using a masking threshold obtained by correction. The masking threshold value correction unit 18 performs a process of setting the upper limit value of the corrected masking threshold value sfbThr (sfb) in each band sfb to a value that does not allow a loss as a loss prevention process (step S22). In order to prevent loss of the band when quantized, the corrected masking threshold value sfbThr (sfb) should not be larger than the power value mdct_pow (sfb). Therefore, in step S22, the upper limit value of the masking threshold value sfbThr (sfb) after correction is set to a value slightly smaller than the power value mdct_pow (sfb).

  Next, the masking threshold value correction unit 18 corrects the masking threshold value based on the auditory characteristic, the upper limit value of the masking threshold value, and the bit rate (step S24). In step S24, for example, the following equation (3) is used to correct the masking threshold so that the PE value calculated based on the power value mdct_pow (sfb) and the corrected masking threshold sfbThr (sfb) becomes the target PE value. To do.

  In Expression (3), r is a correction parameter (see Non-Patent Document 1).

  The first masking threshold correction in step S24 is performed under the condition that there is no band missing when the frequency spectrum is quantized using the corrected masking threshold. When the masking threshold is corrected under such a condition that there is no missing band, the outline of the masking threshold after the correction is greatly deviated from the outline of the initial masking threshold, and the sound quality may be deteriorated. Therefore, in the encoding device 1 of the present embodiment, the approximate similarity of the masking threshold values before and after correction is compared with a predetermined reference value to determine whether or not the corrected masking threshold value is adopted. When the approximate similarity is equal to or lower than the reference value, the masking threshold corrected by the immediately preceding correction process is not adopted, and the masking threshold is corrected again. The recorrection control unit 20 determines whether or not the masking threshold value sfbThr (sfb) after correction is adopted.

After correcting the masking threshold value, the masking threshold value correction unit 18 passes the obtained masking threshold value sfbThr (sfb) to the acceptance / rejection determination unit 20a of the recorrection control unit 20, and stores the upper limit value of the masking threshold value in the storage unit 20c. . Then, the acceptance / rejection determination unit 20a calculates the cross-correlation value corre between the initial masking threshold sfbThr ₀ (sfb) and the corrected masking threshold sfbThr (sfb) in order to determine acceptance / rejection of the corrected masking threshold sfbThr (sfb). (Step S26). Whether or not the corrected masking threshold value sfbThr (sfb) is adopted is determined based on whether or not the calculated cross-correlation value corre is larger than a predetermined reference value (correlation threshold value TH ₁ ) (step S28).

  For example, the threshold outline determining unit 20a calculates the cross-correlation value corre by the following equation (4).

  The cross-correlation value corre is calculated using only the masking threshold of the band to be encoded.

The cross-correlation value corre calculated by Expression (4) is 0 <corre ≦ 1, and the value increases as the similarity of the masking threshold before and after correction increases. Therefore, when the calculated cross-correlation value corre is larger than the correlation threshold TH ₁ (for example, TH ₁ = 0.8) (step S28; Yes), the acceptance / rejection determination unit 20a performs the masking after correction obtained in the immediately preceding correction process. It is determined that the threshold value sfbThr (sfb) is adopted. Then, the acceptance / rejection determination unit 20a passes the masking threshold value sfbThr (sfb) obtained in the immediately preceding correction process to the quantization unit 22. In this case, as illustrated in FIG. 2C, the encoding device 1 quantizes the frequency spectrum using the corrected masking threshold value sfbThr (sfb) (step S34).

On the other hand, if the cross-correlation value corre is less correlation threshold TH ₁ (step S28; No), adoption determination unit 20a does not employ the previous correction process obtained in the corrected masking threshold sfbThr (sfb) (i.e. the masking threshold Is corrected again). In this case, the acceptance / rejection determination unit 20a causes the missing allowance band setting unit 20b to perform a missing allowance band setting process (step S30).

Although omitted in FIG. 2B, when the masking threshold is corrected again, the acceptance / rejection determination unit 20a stores the initial masking threshold sfbThr ₀ (sfb) in the storage unit 20c. The initial masking threshold value sfbThr ₀ (sfb) stored in the storage unit 20c is used to calculate a cross-correlation value corre with the corrected masking threshold value.

In the missing allowable band setting process (step S30), a band that is allowed to be lost when quantized is set. In the present embodiment, each time the process of step S30 is performed, the band with the lowest importance is set as the band that allows the omission, among the bands that have the upper limit value of the masking threshold set to a value that does not allow the omission. The importance of the band is the degree of influence on sound quality degradation when missing due to quantization. In this embodiment, as will be described later, a difference value (that is, a signal-to-mask ratio) between the power value mdct_pow (sfb) and the initial masking threshold value sfbThr ₀ (sfb) is used as the importance of the band. In addition, for the band that allows omission, the upper limit value of the masking threshold is set to a value that can be distinguished from a value that does not allow omission such as “0” or “−1”.

  After completing the setting of the band that allows the loss, the missing allowable band setting unit 20b updates the upper limit value of the masking threshold stored in the storage unit 20c and corrects the control signal instructing the correction of the masking threshold by the masking threshold correction. Send to part 18. Thereby, the missing allowable band setting process (step S30) ends.

When the loss allowable band setting process (step S30) ends, the encoding process performed by the encoding apparatus 1 returns to the process of correcting the masking threshold (step S24). Thereafter, the encoding apparatus 1, until the cross-correlation value corre is greater than the correlation threshold TH _1, and repeats the processing of step S24～S30. When the cross-correlation value corre is greater than the correlation threshold TH _1; performing (step S28 Yes), the quantization using the previous correction processing resulting masking threshold sfbThr (sfb) (step S34).

When the quantization using the initial masking threshold sfbThr ₀ (sfb) (step S32) or the quantization using the corrected masking threshold sfbThr (sfb) (step S34) is completed by the above procedure, the encoding device 1 The quantized value is encoded (step S36). Step S36 is performed by the encoding unit 24.

  The encoding unit 24 performs encoding using a known encoding method such as fixed Huffman encoding. When the encoding is completed, the encoding unit 24 passes the encoded data to the multiplexing unit 26. Thereby, the encoding process for one frame of the input signal (audio data) is completed.

  When the encoding process is completed, the encoding apparatus 1 (multiplexer 26) generates and outputs an encoded stream in which header information and the like are added to the encoded audio data.

As described above, in the encoding process performed by the encoding apparatus 1 according to the present embodiment, when correcting the initial masking threshold, first, all bands sfb are corrected by adding a condition that does not allow missing due to quantization. . If the approximate similarity between the corrected masking threshold sfbThr (sfb) and the initial masking threshold sfbThr ₀ (sfb) is lower than the reference value, the importance is low until the approximate similarity exceeds the reference value. The correction of the masking threshold is repeated while allowing deletion in order from the band. Thereby, it is possible to suppress deterioration in sound quality due to a decrease in approximate similarity (excessive correction) while suppressing deterioration in sound quality due to lack of bands.

  Next, the contents of the missing allowable band setting process (step S30) in the present embodiment will be described.

  FIG. 3 is a flowchart showing the content of the missing allowable band setting process in the first embodiment.

In the present embodiment, as described above, the difference value between the power value mdct_pow (sfb) and the initial masking threshold sfbThr ₀ (sfb) is used as the importance of each band sfb. This difference value is the signal-to-mask ratio calculated in step S20, and is stored in the storage unit 20c before starting the first loss allowable band setting process (step S30). Therefore, when the missing allowable band setting unit 20b performs the missing allowable band setting process, as shown in FIG. 3, first, the signal-to-mask ratio and the upper limit value of the masking threshold value of each band to be encoded are stored from the storage unit 20c. Is read (step S3000).

  Next, the missing allowable band setting unit 20b identifies the band having the lowest signal-to-mask ratio among the bands whose upper limit value of the masking threshold is set to a value that does not allow missing as the band having the lowest importance (step) S3002). The upper limit value of the masking threshold that does not allow omission is a positive value smaller than the power value as described above.

  Next, the missing allowable band setting unit 20b updates the data in the storage unit 20c by changing the upper limit value of the masking threshold for the band with the lowest importance to a value that allows the loss (step S3004). The value that allows the omission may be a value that can be distinguished from the upper limit value of the masking threshold that does not allow the omission, for example, “0” or “−1”.

  When the process of step S3004 is completed, the missing allowable band setting unit 20b ends the missing allowable band setting process (return).

The initial masking threshold sfbThr ₀ (sfb) is a value calculated based on an auditory psychological model, and is a limit value of the quantization error amount that cannot be perceived by humans. A sound in a band where the power value mdct_pow (sfb) in the frequency spectrum is smaller than the initial masking threshold cannot be perceived. Further, even if the power value is larger than the initial masking threshold, it is difficult to perceive a sound in a band where the difference between the two is very small. Therefore, the band having a very small signal-to-mask ratio calculated in step S20 has a smaller influence on the sound quality degradation when missing due to quantization than the band having a large signal-to-mask ratio. Therefore, by using the signal-to-mask ratio as the importance of the band and allowing the loss in order from the band having the smallest signal-to-mask ratio, it is possible to suppress deterioration in sound quality due to the loss of the band.

Furthermore, when the masking threshold is corrected again, if some of the bands are allowed to be missing due to quantization, the masking threshold for the band that allowed the missing can be set to a value larger than the value at the previous correction. Become. If the masking threshold of the band that allows the loss is corrected to a large value, the amount of bits used for encoding the band can be reduced or set to “0”. In this way, if the amount of bits used in a band that allows loss can be reduced, the reduced amount of bits can be used for encoding other bands. Bits obtained by reducing the amount of bits used in a band that allows for loss are allocated to a band in which the difference between the masking threshold values before and after correction is large in the previous correction, for example. Therefore, the corrected masking threshold can be brought close to the initial masking threshold. Therefore, it is possible to suppress deterioration in sound quality due to a deviation between the outline of the masking threshold sfbThr (sfb) used for quantization and the outline of the initial masking threshold sfbThr ₀ (sfb).

  The masking threshold value correction process will be specifically described with reference to FIGS. 4A to 4D.

FIG. 4A is a graph illustrating an example of a frequency spectrum and an initial masking threshold.
When the processing of steps S10 to S14 is performed on the input signal for one frame, for example, the power value mdct_pow (sfb) and the initial masking threshold value sfbThr ₀ (sfb) of each band sfb of the frequency spectrum as shown in FIG. 4A are obtained. can get.

As described above, the initial masking threshold value sfbThr ₀ (sfb) is an optimal masking threshold value for quantization of the corresponding frequency spectrum. Therefore, when it is determined that the quantization using the initial masking threshold sfbThr ₀ (sfb) is possible by the processing in steps S16 and S18, the encoding device 1 uses the initial masking threshold sfbThr ₀ (sfb) to perform frequency spectrum analysis. Is quantized (step S32).

  However, when encoding is performed under a low bit rate condition, that is, when the number of bits that can be used for encoding a frequency spectrum is small, the quantization error often cannot be made lower than the initial masking threshold. When the quantization error cannot be made equal to or less than the initial masking threshold, the encoding apparatus 1 performs correction for increasing (relaxing) the masking threshold within a range where the sound quality is not deteriorated as much as possible based on the bit rate condition, auditory characteristics, and the like.

  FIG. 4B is a graph for explaining the relationship between the masking threshold obtained by the first correction and the initial masking threshold.

  The first correction for the initial masking threshold in the present embodiment is performed under conditions that do not allow missing due to quantization for all bands. When the first correction is performed, the corrected masking threshold value sfbThr (sfb) has, for example, an outline as shown by a solid line in FIG. 4B.

Comparing the outline of the corrected masking threshold sfbThr (sfb) with the outline of the initial masking threshold sfbThr ₀ (sfb) shown by the dotted line in FIG. 4B, the low-frequency bands sfb1 to sfb3 and the high-frequency bands In sfb15 to sfb18, the similarity between the two is low. Still, if the cross-correlation value corre calculated from the masking threshold values sfbThr (sfb) and sfbThr ₀ (sfb) before and after correction is larger than the correlation threshold value TH ₁ , the outline of the masking threshold value is prevented while preventing loss of the band due to quantization. It is possible to suppress deterioration in sound quality due to deviation.

However, before and after correction of the masking threshold sfbThr (sfb), sfbThr ₀ if the cross-correlation value is calculated from (sfb) corre is equal to or less than the correlation threshold value TH _1, some of the band of the masking threshold is excessively corrected by which sound quality Leads to. Thus, the masking threshold sfbThr before and after correction shown in FIG. 4B (sfb), sfbThr0 case cross-correlation value corre calculated from (sfb) is equal to or less than the correlation threshold value TH _1, and allows the omission of the lowest band importance Correct the masking threshold.

  In the present embodiment, the band having the lowest importance is the band having the smallest signal-to-mask ratio. The band with the smallest signal-to-mask ratio in the bands sfb1 to sfb5 and Sfb11 to sfb18 shown in FIG. 4A is the band sfb5. Therefore, the encoding apparatus 1 changes the upper limit value of the masking threshold value of the band sfb5 to a value that allows omission and performs the second correction on the initial masking threshold value.

When the masking threshold value is corrected again under the condition that only the loss of the band sfb5 is allowed, the corrected masking threshold value sfbThr ₂ (sfb) has, for example, an outline as shown by a solid line in FIG. 4C. FIG. 4C is a graph illustrating the relationship between the masking threshold obtained by the second correction and the masking threshold obtained by the first correction.

When the masking threshold value sfbThr ₂ (sfb) obtained by the second correction and the masking threshold value sfbThr (sfb) obtained by the first correction shown by the broken line in FIG. 4C are compared, the second time in several bands. The masking threshold sfbThr ₂ (sfb) obtained by the correction is smaller. This is because the second correction allows the loss of the band sfb5.

In the second correction, the masking threshold sfbThr ₂ (sfb5) for the band sfb5 can be made larger than that in the first correction. If the masking threshold value sfbThr ₂ (sfb5) for the band sfb5 becomes larger than that at the time of the first correction, the bit amount used for coding the band sfb5 is reduced, and is used for coding other bands by the reduced amount. The amount of bits can be increased. If the amount of bits used for encoding one band increases, the masking threshold for that band can be made smaller than that for the first correction. Therefore, in the masking threshold value sfbThr ₂ (sfb) obtained by the second correction, as shown in FIG. 4C, the bands sfb1 and sfb2 in which the difference from the initial masking threshold value sfbThr ₀ (sfb) was large in the first correction. Etc. are smaller than the value at the time of the first correction. That is, in the second correction, since the loss of the band sfb5 is allowed, excessive correction to the masking threshold values of the bands sfb1 and sfb2 is suppressed.

As can be seen from FIG. 4C, 2 nd correction obtained masking threshold sfbThr ₂ (sfb), compared to the first correction is obtained masking threshold sfbThr (sfb), the initial masking threshold sfbThr ₀ (sfb ) And high degree of similarity. Therefore, the cross-correlation value corre between the masking threshold sfbThr ₂ (sfb) obtained by the second correction and the initial masking threshold sfbThr ₀ (sfb) is the initial value of the masking threshold sfbThr (sfb) obtained by the first correction. It becomes larger than the cross-correlation value with the masking threshold value sfbThr ₀ (sfb). When the cross-correlation value corre between the masking threshold value sfbThr ₂ (sfb) obtained by the second correction and the initial masking threshold value sfbThr ₀ (sfb) is larger than the correlation threshold value TH ₁ , the encoding apparatus 1 performs the second time Quantization is performed using the masking threshold value sfbThr ₂ (sfb) obtained by the correction. Thereby, the shift | offset | difference of the masking threshold value used for quantization and the rough form of the initial masking threshold value, in other words, sound quality deterioration due to excessive correction of the masking threshold value can be suppressed. Further, the masking threshold value sfbThr ₂ (sfb) obtained by the second correction shown in FIG. 4C has a masking threshold value equal to or lower than the power value mdct_pow (sfb) in all the bands including the band sfb5 that is allowed to be lost. . Therefore, it is possible to prevent deterioration in sound quality due to lack of bands due to quantization.

However, when the masking threshold sfbThr ₂ (sfb) obtained by the second correction is compared with the initial masking threshold sfbThr ₀ (sfb), the similarity between the high-frequency bands sfb15 to sfb18 is still low. Therefore, depending on the value of the correlation threshold TH _1, sometimes the cross-correlation value corre the masking threshold obtained in the second correction and the initial masking threshold becomes less correlation threshold TH _1. In that case, the encoding apparatus 1 performs the third correction on the initial masking threshold value sfbThr ₀ (sfb). The third correction is performed by additionally setting the band having the lowest importance among the bands that do not allow omission, that is, the band having the next smallest signal-to-mask ratio after the band sfb5. In the bands sfb1 to sfb5 and sfb11 to sfb18 shown in FIG. 4A (FIG. 4C), the band having the smallest signal-to-mask ratio after the band sfb5 is the band sfb15. Therefore, the encoding device 1 performs the third correction on the initial masking threshold value sfbThr ₀ (sfb) in a state where the upper limit value of the masking threshold value of the bands sfb5 and sfb15 is changed to a value that allows the loss.

When the masking threshold value is corrected again under the condition that allows the loss of the bands sfb5 and sfb15, the corrected masking threshold value sfbThr ₃ (sfb) has an approximate shape as shown by a solid line in FIG. 4D. FIG. 4D is a graph for explaining the relationship between the masking threshold obtained by the third correction and the masking threshold obtained by the second correction.

When the masking threshold value sfbThr ₃ (sfb) obtained by the third correction is compared with the masking threshold value sfbThr ₂ (sfb) obtained by the second correction shown by the broken line in FIG. Thus, the masking threshold obtained by the third correction is smaller. This is because the amount of bits used for encoding the other bands sfb1, sfb2, sfb17, sfb18, etc. has increased by the amount of bits used for encoding the band sfb5 that allows loss.

As can be seen from FIGS. 4D and 4C, the masking threshold values sfbThr ₃ (sfb) obtained by the third correction are masking threshold values sfbThr (sfb) and sfbThr ₂ (sfb) obtained by the first and second corrections. Compared to the initial masking threshold value sfbThr ₀ (sfb), the approximate similarity is high. Therefore, the cross-correlation value corre between the masking threshold value sfbThr ₃ (sfb) obtained by the third correction and the initial masking threshold value sfbThr ₀ (sfb) is the masking threshold value obtained by the first and second corrections and the initial masking. It becomes larger than the cross-correlation value with the threshold value sfbThr ₀ (sfb). When the cross-correlation value corre between the masking threshold value sfbThr ₃ (sfb) obtained by the third correction and the initial masking threshold value sfbThr ₀ (sfb) is larger than the correlation threshold value TH ₁ , the encoding apparatus 1 performs the third time Quantization is performed using the masking threshold value sfbThr ₃ (sfb) obtained by the correction. Thereby, the shift | offset | difference of the masking threshold value used for quantization and the rough form of an initial masking threshold value, ie, the sound quality degradation by the excessive correction | amendment of a masking threshold value, can be suppressed.

Further, the masking threshold value sfbThr ₃ (sfb) obtained by the third correction shown in FIG. 4D has a masking threshold value equal to or less than the power value mdct_pow (sfb) in all bands except the band sfb5 that allows the loss. . Accordingly, only the least important band sfb5 is lost due to quantization. Therefore, deterioration in sound quality due to lack of bands can be minimized.

Depending on the value of the correlation threshold TH ₁ , the cross-correlation value corre between the masking threshold sfbThr ₃ (sfb) obtained by the third correction and the initial masking threshold sfbThr ₀ (sfb) may be less than the correlation threshold TH _1. There is also. In that case, the encoding apparatus 1 performs the fourth correction on the initial masking threshold sfbThr ₀ (sfb). In the fourth correction, the band that has the lowest importance among the bands that do not allow omission, that is, the band that has the smallest signal-to-mask ratio (for example, band sfb14) after band sfb15 is additionally set as the band that allows omission. And do it. Thereafter, the encoding apparatus 1, until the cross-correlation value corre between the masking threshold and the initial masking threshold after correction is larger than the correlation threshold TH _1, repeated correction of the masking threshold while adding configure bandwidth to allow missing. Thereby, it is possible to suppress deterioration in sound quality due to excessive correction of the masking threshold while suppressing deterioration in sound quality due to lack of bands.

  The encoding apparatus 1 of the present embodiment that performs the encoding process as described above can be realized by, for example, a computer and a program that causes the computer to execute the encoding process. Hereinafter, the encoding apparatus 1 realized by a computer and a program will be described with reference to FIG.

  FIG. 5 is a schematic diagram illustrating a hardware configuration example of a computer that operates as an encoding apparatus.

  As shown in FIG. 5, the computer 5 that operates as an encoding device includes a Central Processing Unit (CPU) 50, a main storage device 52, an auxiliary storage device 54, an input device 56, and an output device 58. . The computer 5 further includes a digital signal processor (DSP) 60, a storage medium driving device 62, and an interface device 64. These elements 50 to 64 in the computer 5 are connected to each other by a bus 68 so that data can be exchanged between the elements.

  The CPU 50 is an arithmetic processing unit that controls the overall operation of the computer 5 by executing various programs including an operating system.

  The main storage device 52 includes a read only memory (ROM) 52a and a random access memory (RAM) 52b. In the ROM 52a, for example, a predetermined basic control program read by the processor 50 when the computer 5 is started is recorded in advance. The RAM 52b is used as a working storage area as necessary when the processor 50 executes various programs. In the present embodiment, the RAM 52b is used for temporary storage of, for example, an audio signal to be encoded and a masking threshold.

  The auxiliary storage device 54 is a storage device with a larger capacity than the main storage device 52 such as a hard disk drive (HDD) or a solid state disk (SSD). The auxiliary storage device 54 stores various programs executed by the CPU 50, various data, and the like. Examples of the program stored in the auxiliary storage device 54 include an audio player program that performs encoding and reproduction of an audio signal. Further, examples of data stored in the auxiliary storage device 54 include audio signal data encoded by the player.

  The input device 56 is, for example, a keyboard device or a mouse device. When operated by an operator of the computer 5, the input device 56 transmits input information associated with the operation content to the CPU 50.

  The output device 58 is, for example, a liquid crystal display or a speaker. The liquid crystal display displays various texts, images, and the like according to display data transmitted from the CPU 50 or the like. The speaker outputs audio data and audio data transmitted from the CPU 50, DSP 60, and the like.

  The DSP 60 is an arithmetic processing unit that performs audio signal encoding processing, decoding (reproduction) processing, and the like in accordance with control signals from the CPU 50.

  The storage medium driving device 64 reads programs and data recorded in a portable storage medium (not shown) and writes data stored in the auxiliary storage device 54 to the portable storage medium. As the portable storage medium, for example, a flash memory equipped with a USB standard connector can be used. Further, as a portable storage medium, an optical disc such as a Compact Disk (CD), a Digital Versatile Disc (DVD), and a Blu-ray Disc (Blu-ray is a registered trademark) can be used.

  The interface device 64 is, for example, an audio input / output device or a communication control device. The audio input / output device inputs and outputs an audio signal by connecting the computer 5 to a microphone or an audio device, for example. The communication control device communicatively connects the computer 5 and a communication network such as the Internet, and transmits and receives audio data and the like by communication with an external communication device or the like via the communication network.

  In the computer 5, the CPU 50 reads out the program including the encoding process described above from the auxiliary storage device 54, and executes the encoding process of the audio signal in cooperation with the DSP 60, the main storage device 52, the auxiliary storage device 54, and the like. To do. At this time, the CPU 50 causes the DSP 60 to perform arithmetic processing in the encoding process. The DSP 60 converts the audio signal into a frequency spectrum and generates an initial masking threshold. The audio signal may be read from a portable storage medium such as a music CD, for example, or may be input to the computer 5 by communication via the interface device 64. Further, the DSP 60 calculates the initial PE value and the target PE value, and determines whether or not the audio signal can be quantized using the initial masking threshold value based on the magnitude relationship between them. If the initial masking threshold cannot be used for quantization, the DSP 60 calculates an auditory characteristic and corrects the masking threshold. Further, the DSP 60 determines whether or not the corrected masking threshold is adopted, and if not adopted, sets a band that allows omission and corrects the masking threshold again. When it is determined that the corrected masking threshold is adopted, the audio signal (frequency spectrum) is quantized and encoded using the adopted masking threshold. Further, the DSP 60 stores the initial masking threshold value, the target PE value, the upper limit value of the corrected masking threshold value in the RAM 52b and the auxiliary storage device 54, and the RAM 52b and auxiliary storage device during the execution of the above processing. The process of reading from 54 is performed.

  Audio signal data (audio data) encoded by the computer 5 is stored in, for example, the auxiliary storage device 54 and is decoded (reproduced) by the computer 5 as necessary. Further, if the computer 5 includes a communication control device as the interface device 64, for example, audio data can be provided (distributed) to another computer or the like via a communication network.

  The computer 5 used as the encoding device 1 is not limited to the configuration shown in FIG. 5, and may be configured to encode an audio signal in the CPU 50. In addition, the computer 5 used as the encoding device 1 is not limited to a general-purpose computer that realizes a plurality of functions by executing various programs, but may be an audio device specialized for encoding and decoding audio signals. Good.

  As described above, in the encoding method using the encoding device 1 according to the first embodiment, the audio signal (frequency spectrum) cannot be quantized using the initial masking threshold generated based on the psychoacoustic model. The masking threshold is corrected. At this time, the first correction is performed under the condition that all bands are not lost due to quantization, and it is determined whether the approximate similarity of the masking threshold before and after the correction satisfies the reference value. If the approximate similarity does not satisfy the reference value, masking is performed while allowing omissions in order of decreasing importance until the approximate similarity of the masking threshold values before and after correction reaches the reference value. Repeat threshold correction. For this reason, while suppressing deterioration in sound quality due to the loss of some bands due to quantization, due to a shift in approximate similarity between the masking threshold used for quantization and the initial masking threshold (ie, excessive correction of the masking threshold) Sound quality deterioration can be suppressed.

Note that the loss allowable band setting process includes, for example, deleting a plurality of bands in order from the lowest importance band according to the magnitude of the difference between the cross-correlation value corre and the correlation threshold value TH _{1 in one} setting process. It may be allowed.

  Further, whether or not the masking threshold after correction is adopted is not limited to the cross-correlation value corre expressed by the equation (4), but may be determined using a value corresponding to the approximate similarity of the masking threshold before and after correction. .

  Further, the auditory characteristic calculated in step S20 is not limited to the signal-to-mask ratio, and may be another characteristic.

  Furthermore, the encoding apparatus 1 is not limited to an apparatus that only encodes an audio signal as shown in FIG. 1, and may be an apparatus that encodes a video signal. An apparatus for encoding a video signal has a configuration for encoding a moving image in addition to the configuration shown in FIG. In such an apparatus, encoding of an input video signal is divided into moving image encoding and audio encoding, and then the encoded moving image and audio are multiplexed.

[Modification of First Embodiment]
In the masking threshold correction process shown in FIG. 2B, the masking threshold is corrected until the cross-correlation value corre between the corrected masking threshold sfbThr (sfb) and the initial masking threshold sfbThr ₀ (sfb) becomes larger than the correlation threshold TH _1. repeat. However, the encoding process according to the first embodiment is not limited to this. For example, an upper limit may be provided for the number of masking threshold corrections.

  FIG. 6 is a flowchart illustrating a modification of the masking threshold value correction process according to the first embodiment. The correction process shown in FIG. 6 is performed when it is determined in step S18 shown in FIG. 2A that the initial PE value is larger than the target PE value (step S18; Yes).

  When the initial PE value is larger than the target PE value, the encoding apparatus 1 continues to calculate the auditory characteristic based on the frequency spectrum or the like (step S20) and the corrected masking threshold value sfbThr (, as shown in FIG. The upper limit value of sfb) is set (step S22). Thereafter, the encoding device 1 (masking threshold correction unit 18) initializes the correction count M to M = 1 (step S23).

After step S23, the encoding apparatus 1, the correction of the masking threshold (step S24), and and subjected to calculation of the cross-correlation value corre (step S26), the cross-correlation value corre determines greater than the correlation threshold value TH ₁ (step S28). Then, the cross-correlation value corre is greater than the correlation threshold TH _1; quantizing the frequency spectra by using (step S28 Yes), the masking threshold sfbThr obtained by correcting the immediately preceding, as shown in FIG. 2C (sfb) (Step S34).

On the other hand, if the cross-correlation value corre is less correlation threshold TH ₁ (step S28; No), as shown in FIG. 6, it is determined whether the correction count M is the threshold value THM or more (step S29). If M ≧ THM (step S29; Yes), the masking threshold value correction process is terminated, and the frequency spectrum is quantized using the masking threshold value sfbThr (sfb) obtained by the previous correction (step S34). If M <THM (step S29; No), after performing the missing allowable band setting process (step S30), the number of corrections M is increased by 1 (step S31), and the process returns to the correction process of step S24. The threshold THM for the number of corrections may be set as appropriate, for example, about 5 to 10.

  In the masking threshold correction process shown in FIG. 2B, the number of bands that allow omission increases as the number of corrections increases. For this reason, depending on the frequency spectrum of the audio signal, the number of corrections increases and the loss of a highly important band is allowed, and there is a risk of loss due to quantization.

  On the other hand, when an upper limit is set for the number of corrections as shown in FIG. In addition, when an upper limit is set for the number of corrections as shown in FIG. 6, it is possible to suppress delays in encoding processing due to repeated correction of the masking threshold, and AAC-Enhanced Low Delay (AAC-ELD), etc. Can be easily applied to low-delay coding.

  The threshold THM for the number of corrections may be fixed to a specific number, for example, depending on the frequency spectrum pattern, such as the number of bands whose importance (signal-to-mask ratio) is smaller than a predetermined value. You may change it each time.

[Second Embodiment]
In the present embodiment, the masking threshold value correction process in the encoding apparatus according to the present invention can be performed efficiently. The masking threshold value correction process of this embodiment is performed when the initial PE value is larger than the target PE value in the determination of step S18 shown in FIG. 2A (step S18; Yes), as in the first embodiment.

  Note that the encoding apparatus that performs the masking threshold correction process described in the present embodiment includes a part of the processes performed by the masking threshold generation unit 14 and the recorrection control unit 20 in the encoding apparatus 1 illustrated in FIG. It may be changed as follows.

When the initial PE value is larger than the target PE value, the masking threshold value generation unit 14 in the encoding device 1 according to the present embodiment, the initial masking threshold value sfbThr ₀ (sfb), the target PE value, and the power value mdct_pow ( sfb) is passed to the acceptance / rejection determination unit 20a. Further, the masking threshold value generation unit 14 passes the initial masking threshold value sfbThr ₀ (sfb) and the target PE value to the auditory characteristic calculation unit 16. Further, the masking threshold value generation unit 14 stores the initial PE value or the difference value between the target PE value and the initial PE value in the storage unit 20 c of the recorrection control unit 20.

  On the other hand, the re-correction control unit 20 in the encoding device 1 according to the present embodiment performs processing as shown in FIGS.

  FIG. 7 is a flowchart showing a masking threshold correction process according to the second embodiment of the present invention.

  As described above, the masking threshold value correction process according to the present embodiment is performed when the initial PE value is larger than the target PE value (step S18; Yes). When the initial PE value is larger than the target PE value, the frequency spectrum cannot be quantized using the initial masking threshold as described in the first embodiment. Therefore, when the initial PE value is larger than the target PE value, the encoding apparatus 1 corrects the masking threshold according to the procedure shown in FIG.

Also in the present embodiment, when correcting the masking threshold, the auditory characteristic calculator 16 first calculates the auditory characteristic based on the frequency spectrum or the like (step S20). The auditory characteristic calculation unit 16 calculates a difference value (signal-to-mask ratio) between the power value mdct_pow (sfb) of the frequency spectrum and the initial masking threshold sfbThr ₀ (sfb) as the auditory characteristic.

  When the calculation of the auditory characteristic (signal-to-mask ratio) is finished, the masking threshold value correction unit 18 then sets the upper limit value of the corrected masking threshold value sfbThr (sfb) to a value that allows omission (step S21). . When the setting of the upper limit value of the threshold is completed, the masking threshold value correcting unit 18 corrects the masking threshold value based on the auditory characteristics and the upper limit value of the threshold value (step S24). That is, in the present embodiment, the first masking threshold value correction is performed under the condition that the entire band to be encoded is allowed to be lost due to quantization. As in the first embodiment, the masking threshold is calculated using Equation (3), and the PE value calculated based on the power value mdct_pow (sfb) and the corrected masking threshold sfbThr (sfb) is the target PE value. Correct so that

  When the correction of the masking threshold is finished, the acceptance determination unit 20a of the recorrection control unit 20 then compares the corrected masking threshold sfbThr (sfb) with the power value mdct_pow (sfb). Then, the acceptance / rejection determination unit 20a determines whether sfbThr (sfb) <mdct_pow (sfb) is satisfied in all the bands to be encoded (step S27). When sfbThr (sfb) <mdct_pow (sfb) is satisfied in all the bands to be encoded (step S27; Yes), since there is no band missing due to quantization, the acceptance / rejection determination unit 20a determines the corrected masking threshold value sfbThr (sfb ). In this case, as illustrated in FIG. 2C, the encoding apparatus 1 quantizes the frequency spectrum using the masking threshold value sfbThr (sfb) obtained by the previous correction (step S34).

  On the other hand, when there is a band of sfbThr (sfb) ≧ mdct_pow (sfb) (step S27; No), the band is lost when the frequency spectrum is quantized using the corrected masking threshold sfbThr (sfb) related to the determination. . The first correction of the masking threshold is performed under the condition that all the bands are allowed to be lost as described above. Therefore, a band where sfbThr (sfb) ≧ mdct_pow (sfb) may have a high importance regarding sound quality. If a band with high importance is lost due to quantization, the sound quality is significantly degraded. Therefore, if there is a band of sfbThr (sfb) ≥ mdct_pow (sfb), the missing allowable band setting process (step S30) is performed, and the setting of the upper limit value of the masking threshold is changed so as to allow only the missing of the less important band Then, the masking threshold is corrected again. The missing permissible bandwidth setting process is performed by the missing permissible bandwidth setting unit 20b as in the first embodiment.

  FIG. 8 is a flowchart showing the content of the missing allowable band setting process in the second embodiment. FIG. 9 is a graph for explaining a method of determining a ratio of a band that allows loss.

  As shown in FIG. 8, the missing permissible bandwidth setting unit 20b performs the permissible bandwidth setting processing (step S30) of this embodiment on the basis of the difference value between the target PE value and the initial PE value, as shown in FIG. The ratio of the band that allows the loss is determined (step S3020). In step S3020, for example, the initial PE value and the target PE value are read from the storage unit 20c, and the difference value diffPe is calculated. After that, for example, based on the function f (diffPe) shown in FIG. 9, the ratio lack_ratio of the band that allows the loss is determined. The function f (diffPe) is a function having the difference value diffPe as a parameter, and is represented by the following equation (5).

In FIG. 9 and Expression (5), TH ₂ , TH ₃ , and TH ₄ are all arbitrary values, and may be set as appropriate based on the bit rate, the degree of acceptable sound quality degradation, and the like.

  When the ratio of the band that allows the loss is set, the loss allowable band setting unit 20b next reads the signal-to-mask ratio and the upper limit value of the masking threshold of each band to be encoded from the storage unit 20c (step S3022). Subsequently, the loss allowable band setting unit 20b allows the loss to the band having the rank corresponding to the ratio determined in step S3020 in order from the band having the lowest signal-to-mask ratio (importance) (step S3024).

  Thereafter, the missing permissible band setting unit 20b changes the upper limit value of the masking threshold for a band other than the band that allows the loss to a value that does not allow the loss, and updates the data in the storage unit 20c (step S3026). The upper limit value of the masking threshold that does not allow omission is a positive value that satisfies sfbThr (sfb) <mdct_pow (sfb), as described in the first embodiment.

  When the process of step S3026 is completed, the missing allowable band setting unit 20b ends the missing allowable band setting process (return). When the missing allowable band setting process is completed, the process returns to step S24 shown in FIG. 7, and the correction for the initial masking threshold is performed again.

  As described above, in the present embodiment, the first correction with respect to the initial masking threshold is performed under a condition that allows the loss of all bands. Since there is no upper limit value for the masking threshold after correction under the condition that allows the loss of the band, it is possible to prevent excessive correction as described above, and to suppress a decrease in the similarity of the rough shape of the masking threshold before and after correction. . Therefore, if the masking threshold obtained by the first correction is sfbThr (sfb) <mdct_pow (sfb) in all bands, the approximate similarity of the masking threshold before and after correction is such that the sound quality does not deteriorate significantly. It can be said that it is expensive. Therefore, when the masking threshold obtained by the first correction is adopted, the sound quality is not deteriorated due to the lack of the band, and the sound quality deterioration due to the approximate deviation (excessive correction) of the masking threshold is suppressed.

  Further, in the present embodiment, when the masking threshold obtained by the first correction has a band of sfbThr (sfb) ≧ mdct_pow (sfb), it is not important for the sound quality but the band of sfbThr (sfb) ≧ mdct_pow (sfb). The masking threshold value is corrected again by allowing a missing band of a low degree. This suppresses deterioration in sound quality due to a shift in the outline of the masking threshold (excessive correction) while suppressing deterioration in sound quality due to lack of bands.

  Further, in the present embodiment, the masking threshold is corrected again by setting a band that allows omission based on the difference value diffPe between the target PE value and the initial PE value.

  In encoding an audio signal, if the difference value diffPe between the target PE value and the initial PE value increases, the amount of bits deficient during encoding increases. Therefore, when the difference value diffPe is large, the masking threshold correction amount must be increased. However, under conditions that do not allow band loss, the masking threshold for each band cannot be made larger than the upper limit (power value mdct_pow (sfb)). That is, for the band in which the masking threshold reaches the upper limit value in the course of the correction process using Equation (3), the masking threshold cannot be further increased to reduce the amount of bits used for quantization. Therefore, if there is a band whose masking threshold has reached the upper limit, but the number of bits is insufficient, the amount of bits used for encoding by increasing the masking threshold of the band where the masking threshold has not reached the upper limit. Will be reduced. Therefore, when the amount of correction of the masking threshold increases, the degree of similarity between the approximate shape of the masking threshold after correction and the approximate shape of the initial masking threshold decreases under conditions that do not allow omission. As described above, the approximate similarity between the masking threshold value and the initial masking threshold value corrected under a condition that does not allow band loss is indirectly determined from the difference value diffPe between the target PE value and the initial PE value. Can be grasped. That is, before and after correction under conditions that do not allow band loss without performing processing for correcting the masking threshold under conditions that do not allow band loss and processing for calculating the cross-correlation value corre using Equation (4) It is possible to grasp the approximate similarity of the masking thresholds.

As can be seen from the first embodiment, the greater the difference between the approximate similarity of the masking threshold and the predetermined reference value, the greater the number of bands that are allowed to be dropped in order to make the similarity greater than the reference value. . Therefore, in this embodiment, as a function f (diffPe) shown in FIG. 9, becomes greater than the threshold value TH ₂ with the difference value DiffPe the target PE value and the initial PE values missing in proportion to the difference value DiffPe The ratio (number) of allowable bands is increased. However, if the ratio (number) of bands that allow omission is increased, the masking threshold after correction exceeds the upper limit value, and there is a possibility that the band missing due to quantization may increase. Therefore, when the difference value diffPe is larger than another threshold value TH ₃ (> TH ₂ ), the ratio of the band that allows loss is limited to the threshold value TH ₄ .

  As described above, in this embodiment, when the masking threshold is corrected again, the approximate similarity between the corrected masking threshold and the initial masking threshold is determined based on the difference value between the target PE value and the initial PE value. The ratio of the band that allows the loss is set so as to satisfy the reference value. For this reason, it is possible to suppress deterioration in sound quality due to deviation (excessive correction) of the approximate shape of the masking threshold while suppressing deterioration in sound quality due to lack of bands.

  In addition, the calculation amount such as the difference value diffPe between the target PE value and the initial PE value used when setting the band that is allowed to be lost in this embodiment and the importance (signal to mask ratio) of each band is expressed by the equation (4). ) Is less than the amount of calculation of the cross-correlation value corre. In addition, the target PE value, the initial PE value, and the signal-to-mask ratio are calculated by the masking threshold value generation unit 14 and the auditory characteristic calculation unit 16 before the missing allowable band setting process (step S30). Therefore, the masking threshold correction processing according to the present embodiment can reduce the amount of calculation compared to the first embodiment in which the cross-correlation value corre of the masking threshold is calculated to determine the approximate similarity. Therefore, according to the present embodiment, the masking threshold value correction process can be performed efficiently.

  Furthermore, in the masking threshold correction process according to the present embodiment, when the difference value diffPe between the target PE value and the initial PE value is large, the missing of two or more bands having different degrees of importance in one missing permissible band setting process. Can be tolerated. Therefore, the correction process when correcting the masking threshold can be performed more efficiently, and the application to low delay encoding such as AAC-Enhanced Low Delay (AAC-ELD) becomes easier.

  In the present embodiment, the ratio lack_ratio of the band that allows missing when correcting the masking threshold is determined based on the function f (diffPe) represented by FIG. 9 and Expression (5). However, the function f (diffPe) is not limited to this. For example, an arbitrary sigmoid function may be used.

[Third Embodiment]
In the present embodiment, a band that allows omission is set based on a value different from that of the second embodiment, so that the masking threshold correction process in the encoding device according to the present invention can be performed efficiently. is there. The masking threshold value correction process of this embodiment is performed when the initial PE value is larger than the target PE value in the determination of step S18 shown in FIG. 2A (step S18; Yes), as in the first embodiment. Further, the masking threshold correction processing according to the present embodiment is performed according to the procedure shown in FIG. That is, in the masking threshold value correction process of the present embodiment, the first correction for the initial masking threshold value is performed under a condition that allows the loss of all bands (steps S21 and S24). If there is a band where the relationship between the power value mdct_pow (sfb) of the frequency spectrum and the corrected masking threshold value sfbThr (sfb) is sfbThr (sfb) ≧ mdct_pow (sfb) (step S27; Yes), the masking threshold value Correct again. When the masking threshold is corrected again, a loss allowable band setting process (step S30) is performed to set a band that allows a loss due to quantization.

  Note that the encoding apparatus that performs the masking threshold correction process described in the present embodiment includes a part of the processes performed by the masking threshold generation unit 14 and the recorrection control unit 20 in the encoding apparatus 1 illustrated in FIG. It may be changed as follows.

When the initial PE value is larger than the target PE value, the masking threshold value generation unit 14 in the encoding device 1 according to the present embodiment, the initial masking threshold value sfbThr ₀ (sfb), the target PE value, and the power value mdct_pow ( sfb) is passed to the acceptance / rejection determination unit 20a. Further, the masking threshold value generation unit 14 passes the initial masking threshold value sfbThr ₀ (sfb) and the target PE value to the auditory characteristic calculation unit 16.

  Further, the re-correction control unit 20 in the encoding device 1 according to the present embodiment performs the masking threshold value correction process shown in FIG. 7 as described above, but as the missing allowable band setting process (step S30), FIG. The process shown in

  FIG. 10 is a flowchart showing the content of the missing allowable band setting process according to the third embodiment of the present invention. FIG. 11 is a graph for explaining a method of determining a ratio of a band that allows loss.

  The missing permissible band setting process shown in FIG. 10 is performed by the missing permissible band setting unit 20b of the recorrection control unit 20. First, the tolerable band setting unit 20b determines the ratio of bands that allow for loss based on the ratio of the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb) to the number of bands to be encoded (step S3021). ). In step S3021, for example, based on the function f (sat_ratio) shown in FIG. The function f (sat_ratio) is a function whose parameter is the ratio of the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb) to the number of bands to be encoded, and is expressed by the following equation (6).

  In equation (6), encode_sfb_num is the number of bands to be encoded. Further, sat_sfb_num is the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb).

Each of TH ₅ , TH ₆ , and TH ₇ in FIG. 11 and Expression (6) is an arbitrary value, and may be set as appropriate based on the bit rate, the degree of acceptable sound quality degradation, and the like.

  When the ratio of the band that allows the loss is set, the loss allowable band setting unit 20b next reads the signal-to-mask ratio and the upper limit value of the masking threshold of each band to be encoded from the storage unit 20c (step S3022). Subsequently, the loss allowable band setting unit 20b allows the loss to the band of the rank corresponding to the ratio determined in step S3021 in order from the band having the lowest signal-to-mask ratio (importance) (step S3024).

  Thereafter, the missing permissible band setting unit 20b changes the upper limit value of the masking threshold for a band other than the band that allows the loss to a value that does not allow the loss, and updates the data in the storage unit 20c (step S3026). The upper limit value of the masking threshold that does not allow omission is a positive value that satisfies sfbThr (sfb) <mdct_pow (sfb), as described in the first embodiment.

  When the process of step S3026 is completed, the missing allowable band setting unit 20b ends the missing allowable band setting process (return). When the missing allowable band setting process is completed, the process returns to step S24 shown in FIG. 7, and the correction for the initial masking threshold is performed again.

  As described above, in the present embodiment, the first correction with respect to the initial masking threshold is performed under a condition that allows the loss of all bands. In the masking threshold correction process under the condition that allows the loss of the band, a decrease in the similarity of the approximate shape of the masking threshold before and after the correction is suppressed. Therefore, if the masking threshold obtained by the first correction is sfbThr (sfb) <mdct_pow (sfb) in all bands, the approximate similarity of the masking threshold before and after correction is such that the sound quality does not deteriorate significantly. It can be said that it is expensive. Therefore, when the masking threshold obtained by the first correction is adopted, the sound quality is not deteriorated due to the lack of the band, and the sound quality deterioration due to the approximate deviation (excessive correction) of the masking threshold is suppressed.

  Also in the present embodiment, when the corrected masking threshold includes a band of sfbThr (sfb) ≧ mdct_pow (sfb), the band that allows missing is set and the masking threshold is corrected again. At this time, the band that allows the loss is set to allow the loss in order from the lowest importance band based on the ratio of the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb) to the number of bands to be encoded. To do. This suppresses deterioration in sound quality due to a shift in the outline of the masking threshold (excessive correction) while suppressing deterioration in sound quality due to lack of bands.

  When there is a band of sfbThr (sfb) ≧ mdct_pow (sfb) in the masking threshold used for quantization, the band is lost due to quantization. Therefore, in order to suppress the loss of the band due to quantization, it is necessary to correct the band masking threshold of sfbThr (sfb) ≧ mdct_pow (sfb) so that sfbThr (sfb) <mdct_pow (sfb). That is, the amount of bits used for quantization of the band must be increased so that the masking threshold of the band of sfbThr (sfb) ≧ mdct_pow (sfb) becomes small. However, the number of bits that can be used for quantization of the frequency spectrum has an upper limit corresponding to the bit rate. Therefore, in order to reduce the masking threshold of the band of sfbThr (sfb) ≧ mdct_pow (sfb), the masking threshold of the other band of sfbThr (sfb) <mdct_pow (sfb) must be increased. Therefore, if the masking threshold corrected under the conditions that allow for omissions is large in the number of bands that satisfy sfbThr (sfb) ≥ mdct_pow (sfb), The approximate similarity with the initial masking threshold is lowered. That is, the approximate similarity between the masking threshold corrected under the condition not allowing the loss of the band and the initial masking threshold is the band where sfbThr (sfb) ≧ mdct_pow (sfb) in the masking threshold corrected under the condition allowing the loss. It can be indirectly grasped from the number.

As can be seen from the first embodiment, the greater the difference between the approximate similarity of the masking threshold and the predetermined reference value, the greater the number of bands that are allowed to be dropped in order to make the similarity greater than the reference value. . Therefore, in this embodiment, as a function f (sat_ratio) shown in FIG. 11, the missing proportion to the ratio Sat_ratio the sfbThr (sfb) ≧ mdct_pow (sfb ) to become the ratio of the bandwidth Sat_ratio is greater than the threshold value TH ₅ The ratio (number) of bands that allow the increase is increased. However, if the ratio (number) of bands that allow omission is increased, the masking threshold after correction exceeds the upper limit value, and there is a possibility that the band that is lost due to quantization may increase. Therefore, when the ratio sat_ratio of the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb) to the number of bands to be encoded is larger than another threshold TH ₆ (> TH ₅ ), the band that allows the loss limiting the rate of the threshold value TH _7.

  As described above, in this embodiment, when the masking threshold is corrected again, the band with low importance is based on the ratio of the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb) to the number of bands to be encoded. Missing is allowed in order. Then, the correction of the masking threshold is repeated while adding a band that allows omission until the approximate similarity between the corrected masking threshold and the initial masking threshold satisfies the reference value. For this reason, it is possible to suppress deterioration in sound quality due to deviation (excessive correction) of the approximate shape of the masking threshold while suppressing deterioration in sound quality due to lack of bands.

  Further, the values used when setting the band that allows the loss in this embodiment are the masking threshold value generation unit 14, the auditory characteristic calculation unit 16, and the masking threshold value correction before the deletion allowable band setting process (step S 30). This is calculated by the unit 18. Therefore, the masking threshold correction processing according to the present embodiment can reduce the amount of calculation compared to the first embodiment in which the cross-correlation value corre of the masking threshold is calculated to determine the approximate similarity. Therefore, according to the present embodiment, the masking threshold value correction process can be performed efficiently.

  Furthermore, in the masking threshold correction process according to the present embodiment, when the ratio of the number of bands of sfbThr (sfb) ≧ mdct_pow (sfb) to the number of bands to be encoded is large, it is important in one missing band setting process Missing of two or more bands having different degrees can be allowed. Therefore, the correction process when correcting the masking threshold can be performed more efficiently, and the application to low-delay encoding such as AAC-ELD becomes easier.

  In this embodiment, based on the function f (sat_ratio) represented by FIG. 11 and the equation (6), the ratio lack_ratio of the band that allows omission when correcting the masking threshold is determined. However, the function f (sat_ratio) is not limited to this, and an arbitrary sigmoid function may be used, for example.

The following additional notes are further disclosed with respect to the embodiments including the examples described above.
(Appendix 1)
In an encoding device that converts an audio signal into a frequency spectrum and performs quantization and encoding of the frequency spectrum.
A threshold generation unit that generates an initial masking threshold when quantizing based on the frequency spectrum;
A threshold correction unit that corrects the initial masking threshold based on the amount of bits available for quantization of the frequency spectrum;
Sets whether or not to allow omission due to quantization for each band of the frequency spectrum when the approximate similarity between the masking threshold corrected by the threshold correction unit and the initial masking threshold is equal to or less than a reference value. A re-correction control unit that causes the threshold correction unit to correct the initial masking threshold again;
An encoding device comprising:
(Appendix 2)
The re-correction control unit again sets the initial masking threshold to the threshold correction unit when the similarity between the initial masking threshold and the corrected masking threshold that does not allow omission of the approximate shape of the band is equal to or less than the reference value. To correct,
The encoding device according to appendix 1, wherein
(Appendix 3)
The re-correction control unit calculates a cross-correlation value between the initial masking threshold and the masking threshold corrected by the threshold correction unit as the approximate similarity of the masking threshold.
The encoding apparatus according to Supplementary Note 2, wherein
(Appendix 4)
The re-correction control unit determines a ratio of bands that allow a loss with respect to the entire band to be encoded based on the approximate similarity between the initial masking threshold and the corrected masking threshold, and the determined ratio A setting unit that sets whether or not to allow missing for each band of the frequency spectrum based on
The encoding device according to appendix 1, wherein
(Appendix 5)
The setting unit includes a first perceptual entropy value based on the frequency spectrum and the initial masking threshold as the approximate similarity of the masking threshold, and a second based on the number of bits usable for quantization of the frequency spectrum. Using the difference value with the perceptual entropy value, and determining the ratio based on the difference value;
The encoding apparatus according to supplementary note 4, wherein
(Appendix 6)
The setting unit uses a ratio of the number of bands whose corrected masking threshold is larger than the power value of the frequency spectrum to the number of bands to be encoded as the approximate similarity of the masking threshold. A ratio of a band that allows a loss to the entire band to be encoded is determined based on
The encoding apparatus according to supplementary note 4, wherein
(Appendix 7)
The re-correction control unit allows missing in order from a less important band of the frequency spectrum band,
The encoding device according to appendix 1, wherein
(Appendix 8)
The recorrection control unit uses a difference value between the power value in each band and the initial masking threshold as the importance.
The encoding apparatus according to appendix 7, characterized by:
(Appendix 9)
The re-correction control unit employs the corrected masking threshold regardless of the approximate similarity between the corrected masking threshold and the initial masking threshold when the number of corrections with respect to the initial masking threshold reaches a predetermined number. ,
The encoding device according to appendix 1, wherein
(Appendix 10)
The threshold generation unit generates the initial masking threshold based on an auditory psychological model.
The encoding device according to appendix 1, wherein
(Appendix 11)
Computer
When the initial masking threshold generated based on the frequency spectrum obtained from the audio signal does not satisfy the condition for quantizing the frequency spectrum,
Set whether to allow missing due to quantization for each band of the frequency spectrum,
Correcting the initial masking threshold based on the amount of bits available for quantization of the frequency spectrum and whether to allow missing of each band;
If the approximate similarity between the corrected masking threshold and the initial masking threshold is less than or equal to the reference value, the initial masking threshold is corrected again by changing the setting as to whether or not one or more bands are missing. ,
An encoding method characterized by executing processing.
(Appendix 12)
The computer calculates a cross-correlation value between the corrected masking threshold and the initial masking threshold as the approximate similarity.
The encoding method according to attachment 11, wherein the processing is executed.
(Appendix 13)
The computer is
Calculating a difference value between a first perceptual entropy value based on the frequency spectrum and the initial masking threshold and a second perceptual entropy value based on the number of bits usable for quantization of the frequency spectrum;
Based on the calculated difference value, a ratio of a band that allows a loss to a band to be encoded is determined.
The encoding method according to attachment 11, wherein the processing is executed.
(Appendix 14)
The computer is
Based on the ratio of the number of bands whose corrected masking threshold is larger than the power value of the frequency spectrum to the number of bands to be encoded, the ratio of the bands that are allowed to be missing from the band to be encoded is determined. ,
The encoding method according to attachment 11, wherein the processing is executed.
(Appendix 15)
When the corrected masking threshold does not satisfy a predetermined adoption condition, the loss is allowed in order from the band in which the difference value between the power value and the initial masking threshold is small in each band of the frequency spectrum.
The encoding method according to supplementary note 11, wherein
(Appendix 16)
When the initial masking threshold generated based on the frequency spectrum obtained from the audio signal does not satisfy the condition for quantizing the frequency spectrum,
Set whether to allow missing due to quantization for each band of the frequency spectrum,
Correcting the initial masking threshold based on the amount of bits available for quantization of the frequency spectrum and whether to allow missing of each band;
If the approximate similarity between the corrected masking threshold and the initial masking threshold is less than or equal to the reference value, the setting of whether or not one or more missing bands are allowed is changed and the initial masking threshold is corrected again;
A program that causes a computer to execute processing.

DESCRIPTION OF SYMBOLS 1 Encoding apparatus 10 Block switching part 12 MDCT process part 14 Masking threshold value production | generation part 16 Auditory characteristic calculation part 18 Masking threshold value correction | amendment part 20 Recorrection control part 20a Acceptance determination part 20b Missing allowance band setting part 20c Storage part 22 Quantization part 24 Encoder 26 Multiplexer 5 Computer 50 CPU
52 Main storage device 52a ROM
52b RAM
54 Auxiliary storage device 56 Input device 58 Output device 60 DSP
62 Storage medium drive device 64 Interface device

Claims (8)

In an encoding device that converts an audio signal into a frequency spectrum and performs quantization and encoding of the frequency spectrum.
A threshold generation unit that generates an initial masking threshold when quantizing based on the frequency spectrum;
A threshold correction unit that corrects the initial masking threshold based on the amount of bits available for quantization of the frequency spectrum;
Sets whether or not to allow omission due to quantization for each band of the frequency spectrum when the approximate similarity between the masking threshold corrected by the threshold correction unit and the initial masking threshold is equal to or less than a reference value. A re-correction control unit that causes the threshold correction unit to correct the initial masking threshold again;
An encoding device comprising: The re-correction control unit again sets the initial masking threshold to the threshold correction unit when the similarity between the initial masking threshold and the corrected masking threshold that does not allow omission of the approximate shape of the band is equal to or less than the reference value. To correct,
The encoding apparatus according to claim 1. The re-correction control unit determines a ratio of bands that allow a loss with respect to the entire band to be encoded based on the approximate similarity between the initial masking threshold and the corrected masking threshold, and the determined ratio A setting unit that sets whether or not to allow missing for each band of the frequency spectrum based on
The encoding apparatus according to claim 1. The setting unit includes a first perceptual entropy value based on the frequency spectrum and the initial masking threshold as the approximate similarity of the masking threshold, and a second based on the number of bits usable for quantization of the frequency spectrum. Using the difference value with the perceptual entropy value, and determining the ratio based on the difference value;
The encoding apparatus according to claim 3. The setting unit uses a ratio of the number of bands whose corrected masking threshold is larger than the power value of the frequency spectrum to the number of bands to be encoded as the approximate similarity of the masking threshold. A ratio of a band that allows a loss to the entire band to be encoded is determined based on
The encoding apparatus according to claim 3. The re-correction control unit allows missing in order from a less important band of the frequency spectrum band,
The encoding apparatus according to claim 1. Computer
When the initial masking threshold generated based on the frequency spectrum obtained from the audio signal does not satisfy the conditions for use in quantization of the frequency spectrum,
Set whether to allow missing due to quantization for each band of the frequency spectrum,
Correcting the initial masking threshold based on the amount of bits available for quantization of the frequency spectrum and whether to allow missing of each band;
If the approximate similarity between the corrected masking threshold and the initial masking threshold is less than or equal to a reference value, the initial masking threshold is corrected again by changing the setting of whether one or more bands are missing. To
An encoding method characterized by executing processing. When the initial masking threshold generated based on the frequency spectrum obtained from the audio signal does not satisfy the conditions for use in quantization of the frequency spectrum,
Set whether to allow missing due to quantization for each band of the frequency spectrum,
Correcting the initial masking threshold based on the amount of bits available for quantization of the frequency spectrum and whether to allow missing of each band;
If the approximate similarity between the corrected masking threshold and the initial masking threshold is less than or equal to a reference value, the initial masking threshold is corrected again by changing the setting of whether one or more bands are missing. To
A program that causes a computer to execute processing.

Images