JP2002300042A - Signal decoding method, signal decoder, and recording medium having signal decoding processing program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To conduct nonlinear inverse quantization processing with high accuracy and a small arithmetic amount in the case of decoding data that are converted into frequency region data, quantized, encoded and compressed. SOLUTION: The signal decoding method of this invention includes a step s1 of dividing decoded input data X into higher-order bits X- high and lower- order bits X- low, a step s2 of checking whether the higher-order bits are 0, a step s3 of referencing an inverse quantization table Q by using the lower-order bits for a Key when the higher-order bits are 0 to obtain a result of the inverse quantization arithmetic operation, a step s4 of referencing a representative table V to obtain a representative value V by using the higher-order bits for the Key when the higher-order bits are not 0 and referencing an interpolation table G by using the higher-order bits for the Key to obtain the gradient g at that time, and a step s5 of using the representative value V, the gradient g and the divided lower-order bits to calculate V+g*X- low to obtain the result of the nonlinear inverse quantization arithmetic operation with respect to the input data X at that time.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal decoding method, a signal decoding apparatus, and a recording medium on which a signal decoding processing program is recorded. More particularly, the present invention relates to an audio signal such as music or voice converted to a frequency domain and then quantized. The present invention relates to a signal decoding method and a signal decoding device for decoding encoded and compressed data, and a recording medium on which a signal decoding processing program is recorded.

[0002]

2. Description of the Related Art Various techniques for efficiently encoding an audio signal have been proposed. Audio signal 2
Recently, signals having a band of about 0 KHZ have been
(MotionPicture Experts Group) Encoded using an audio system or the like.

In an encoding system represented by the MPEG audio system, a digital audio signal on a time axis is converted into a frequency axis by using an orthogonal transform, and information on the frequency axis is considered in consideration of sensitivity characteristics of human hearing. Then, non-linear quantization is performed, in which information that is important to human hearing is given priority to the amount of information. This takes advantage of the fact that human hearing is sensitive to small sound changes but less sensitive to loud changes in sound. Is a process of assigning a large amount of information to, and reducing the amount of information to a loud sound.

[0004] Further, after performing the variable length coding on the result of the nonlinear quantization, the coded data is output as compressed data of a bit stream.

[0005] With regard to the MPEG audio system, for example, ISO / IEC standard ISO-11172
-3 and ISO13818-3.

[0006]

The decoding of compressed data encoded using the above-described method is performed in the reverse order of the encoding. That is, the variable-length code data of the bit stream is decoded by variable-length code decoding means, nonlinearly dequantized by inverse quantum processing means, and then converted from frequency axis to time-axis data by frequency / time conversion means. To play as an audio signal.

[0007] By the way, the original audio signal is a signal having a resolution of at most about 16 bits on the time axis, but when this signal is converted to the frequency axis, its dynamic range becomes very large. For example, when a signal having a resolution of about 16 bits is simply converted to a frequency axis, the information cannot be expressed unless about 32 bits (ideally, 80 bits or more) are used.

Therefore, as described above, in the MPEG audio system, non-linear quantization is performed in the process of coding, and the result is further subjected to variable-length coding to compress data. The non-linear quantization process performed by the MPEG audio system is based on the ISO / IEC standard I described above.
It is represented by the following equation shown in SO-11172-3.

[0009]

(Equation 1)

By performing the non-linear quantization processing using the equation (1), the compression rate is improved while minimizing the influence on human hearing.

However, as a price, when performing the non-linear quantization or the non-linear inverse quantization, it is necessary to perform a floating-point operation using the above-mentioned equation, and the amount of operation becomes extremely large. Therefore, when performing such quantization or inverse quantization, hardware having a high processing capability corresponding to the quantization or inverse quantization is required.

Note that in the equation (1), is (i)
Since the calculation term following the calculation term of the fourth power of is a constant, the value of the constant part can be obtained in advance.
Since there is no need to perform the calculation each time, the influence on the amount of calculation is small, but the calculation term of the fourth power of is (i) needs to be calculated each time for the input data. . In particular, calculations such as non-integer powers (rational powers) require a large amount of calculation, so the calculation load becomes very large, hardware with high processing capability is required, and the calculation cost becomes very high.

In particular, portable audio equipment for reproducing such compressed data is required to be small, light and low in cost, so that the capacity of the CPU to be used is generally greatly restricted. is there.

Therefore, as a method for reducing the operation load of such nonlinear inverse quantization, the above-mentioned equation (1) is previously calculated for all input data using a computer having a high processing capability. There is a method in which an inverse quantization table storing a result of a nonlinear inverse quantization operation for each input data that may be input is created, and the inverse quantization table is stored in a ROM or the like of an audio device. .

According to this, input data is converted to Ke
By simply referring to the inverse quantization table as y, the nonlinear inverse quantization operation result to be obtained can be obtained, and the operation load can be greatly reduced.However, if the data range that can be input is large, the inverse The data amount of the quantization table becomes enormous, and the ROM for storing the data has to use a large storage capacity.

By the way, if the data input to the inverse quantization means is 16 bits, it is necessary to have the result of the nonlinear inverse quantization operation for 2 @ 16 input data. Occupies a lot of areas.

As described above, portable audio equipment and the like are often required to be small, lightweight, and low in cost, and it is desirable that the capacity of the ROM be as small as possible. It is impractical to use a large-capacity ROM to reduce the amount of computation.

Accordingly, an object of the present invention is to realize the above-described nonlinear inverse quantization operation processing at high speed and with high accuracy by using inexpensive hardware.

[0019]

In order to achieve the above-mentioned object, a signal decoding method according to the present invention converts data into data on a frequency axis, and then performs non-linear quantization according to the required degree of accuracy. Further, the compressed data compressed by being encoded is used as input data, and the compressed data as input data is decoded, and then subjected to nonlinear inverse quantization according to the required degree of accuracy, and then the data on the time axis is decoded. A signal decoding method for converting the input data having a high degree of accuracy demand into each of the decoded input data.
An inverse quantization table capable of acquiring a nonlinear inverse quantization operation result corresponding to each compressed data is prepared as a key. For the input data having a low accuracy requirement, each input at a predetermined interval decoded is provided. A representative value table is prepared which can obtain, as a representative value, the result of the non-linear inverse quantization operation corresponding to the input data at each of the predetermined intervals using the data as a key, and an interpolation table capable of interpolating between these representative values. Then, when performing the non-linear inverse quantization process according to the required degree of accuracy, the required degree of accuracy of the decoded input data is determined. Also, when it is determined that the input data has a high degree of accuracy requirement, by referring to the inverse quantization table using the input data as a key, a nonlinear inverse quantization operation result for the input data is obtained. If it is determined that the input data requires a low degree of accuracy, the input data is referred to as a key to the representative value table or to both the representative value table and the interpolation table, and the nonlinearity to the input data is determined. An inverse quantization operation result is obtained.

Further, the signal decoding apparatus of the present invention converts the compressed data which has been compressed by being converted into data on the frequency axis, nonlinearly quantized in accordance with the required degree of accuracy, and further encoded. A signal decoding device that decodes compressed data as input data, decodes the compressed data as input data, performs a non-linear inverse quantization process in accordance with the degree of accuracy required, and converts it into data on a time axis. The apparatus includes a decoding unit for decoding the encoded input data, an inverse quantization unit for performing non-linear inverse quantization on the decoded data, and converting the non-linear inverse-quantized data into time-axis data. The frequency / time conversion means for conversion is included as a component. Then, the inverse quantization means, as a table for the input data having a high degree of accuracy requirement, can perform inverse quantization that can obtain a nonlinear inverse quantization operation result corresponding to each input data by using each decoded input data as a key. Having a decoding table, and as a table for the input data having a low degree of accuracy requirement, using the decoded input data at each predetermined interval as a key, a non-linear inverse quantization operation result corresponding to the input data at the predetermined interval. It has a representative value table that can be obtained as a representative value,
It has an interpolation table that can interpolate between these representative values. In such a configuration, when performing the nonlinear inverse quantization process according to the required degree of accuracy, the required degree of accuracy of the decoded input data is determined, and the input data having a high required degree of accuracy is determined. If determined, by referring to the inverse quantization table with the input data as Key,
A non-linear inverse quantization operation result for the input data is obtained. When it is determined that the input data has a low accuracy requirement, by referring to the representative value table or both the representative value table and the interpolation table with the input data as a key, the input data is determined. A non-linear inverse quantization operation result is obtained.

The recording medium on which the signal decoding processing program of the present invention is recorded is converted into data on the frequency axis, then subjected to non-linear quantization according to the required accuracy, and further encoded. A signal decoding processing program that uses compressed compressed data as input data, decodes the compressed data as input data, performs nonlinear inverse quantization according to the degree of accuracy required, and converts it to time-axis data. A non-linear inverse quantization process according to the required degree of accuracy, the processing program determines the required degree of accuracy of the decoded input data, If it is determined that the input data has a high degree of request, inverse quantization can be used to obtain the nonlinear inverse quantization operation result for each input data using the input data as a key. A procedure for referring to the input data and obtaining a result of the non-linear inverse quantization operation on the input data; and, when it is determined that the input data is compressed data having a low degree of required accuracy, the input of each input at a predetermined interval. Refer to the representative value table that can obtain the non-linear inverse quantization operation result corresponding to the input data at the predetermined interval using the data as the key, or both the representative value table and the interpolation table that can interpolate between the representative values. Obtaining a non-linear inverse quantization operation result with respect to the input data.

In each of these inventions, the determination of the required degree of accuracy of the input data is performed by assuming that the number of bits of the decoded input data is n (n is a positive integer), and The number is divided into n1) and lower bits (the number of bits is n2, n1 + n2 = n).
It is determined whether or not the data of the upper bit is 0. If the data of the upper bit is 0, the degree of accuracy required for the input data is high, and if the data of the upper bit is not 0, It is determined that the required degree of accuracy of the input data is low.

In the inverse quantization table, a non-linear inverse quantization operation result calculated in advance for each data in the data range that the lower bits can take is associated with each data that the lower bits can take. Wherein the representative value table stores the non-linear inverse quantization operation result calculated in advance for each possible data of the upper bits for each possible data range interval of the lower bits. A table stored in association with each possible data of the upper bit side, wherein the interpolation table is a non-linear inverse of each possible data of the upper bit for each possible data range of the lower bit. The slope value of the line segment connecting the quantization operation results with the corresponding data of the upper bit Advance determined in advance calculated, and its gradient value is stored in association with each of the data may take the upper bits table.

The processing for obtaining the result of the non-linear inverse quantization operation on the input data when it is determined that the input data requires a high degree of accuracy is performed by converting the data represented by the lower bits of the input data into a key. Is a process of referring to the inverse quantization table and obtaining a non-linear inverse quantization operation result calculated in advance from the inverse quantization table.

On the other hand, the processing for obtaining the result of the non-linear inverse quantization operation on the input data when it is determined that the input data has a low degree of accuracy is performed by converting the data represented by the upper bits of the input data into a key. Refer to the representative value table as above, obtain a non-linear inverse quantization operation result calculated in advance from the representative value table, and when interpolation corresponding to data of lower bits following the upper bits is required, After obtaining the non-linear inverse quantization operation result calculated in advance from the value table, referring to the interpolation table with the data of the upper bits as a key to obtain a slope value corresponding to the upper bits, from the slope value An interpolation value corresponding to the data of the lower bit following the upper bit is obtained, and the interpolation value is calculated by the nonlinear inverse quantization operation obtained from the representative value table. In addition to the results, and to obtain a non-linear inverse quantization operation results.

When it is determined that the input data has a low degree of accuracy, the result of the non-linear inverse quantization operation on the input data is obtained by calculating the non-linear inverse quantization operation result to be obtained from Vx and the representative value table. When the obtained nonlinear inverse quantization operation result is represented by V, the gradient value obtained from the interpolation table is represented by g, and the data of the lower bit following the upper bit is represented by X_low, Vx = V + g * X_low is obtained. I have.

As described above, according to the present invention, when performing the non-linear inverse quantization processing according to the degree of accuracy required in the decoding processing of compressed data, the input compressed data (input data)
Is determined, and if the input data has a high accuracy requirement, a non-linear inverse quantization operation result for the input data is obtained by referring to a previously created inverse quantization table, and the accuracy is determined. If it is determined that the input data has low demand, the nonlinear dequantization result for the compressed data is obtained by referring to the representative value table or both the representative value table and the interpolation table.

As described above, for input data for which a high degree of accuracy is required, the result of the nonlinear inverse quantization operation is obtained using a table capable of obtaining the result of the nonlinear inverse quantization operation for each data. For the input data of low demand, the result of the nonlinear inverse quantization operation is obtained by referring to the representative value table or both the representative value table and the interpolation table.
Compared to calculating the calculation term of (i) to the fourth power as it is, the calculation amount can be greatly reduced, and the CPU
Can be significantly reduced.

Further, as compared with a case where an inverse quantization table corresponding to all input data in a data range that can be input is provided, all of the inverse quantization table, the representative value table, and the interpolation table used in the present invention are used. Even in total, the data amount can be significantly reduced, and the capacity of the memory for storing the table can be reduced.

As a result, the present invention enables high-speed signal decoding with inexpensive hardware.
For data requiring high accuracy, the result of the nonlinear inverse quantization operation corresponding to each data can be obtained as it is, so that high-precision decoding processing can be realized. Therefore,
When the signal to be processed is, for example, an audio signal, a high-quality reproduced sound can be obtained, and it can be realized with inexpensive hardware, so that it can be easily mounted on a portable inexpensive device.

The accuracy of the input data is determined by determining whether or not the data on the upper bit side of the input data is 0. If the data on the upper bit side is 0, the input level is determined. If the accuracy of the data is high and the data of the upper bit is not 0, it is determined that the accuracy of the input data is low.

This is based on the fact that, in the case of an audio signal, the input data to be subjected to nonlinear inverse quantization is based on the fact that human hearing is sensitive to small sound changes but not so sensitive to large sound changes. It is determined whether the value of (decoded data) is large or small, thereby determining the required degree of accuracy of the input data. In the present invention, the input data is divided into upper bits and lower bits, and if the data of the upper bits is 0, it is determined that the value of the input data is small and the degree of accuracy required is high. If the data is not 0, the value of the input data is determined to be large and the required degree of accuracy is determined to be low, so that the required degree of accuracy can be properly determined by simple processing.

The inverse quantization table used in the present invention is characterized in that the non-linear inverse quantization operation result calculated in advance for each data in the data range that the lower bits can take corresponds to the individual data that the lower bits can take. This is a table that is attached and stored.

Thus, for example, when the lower bit is 8
In the case of bits, the table may store a non-linear inverse quantization operation result corresponding to 2 8 (256) input data.

Further, the representative value table indicates that the result of the non-linear inverse quantization operation calculated in advance for each possible data of the upper bit for each possible data range interval of the lower bit is the data of the upper bit. It is a table stored in association with individual data to be obtained.

As described above, since the representative value table is a table corresponding to each possible data of the upper bits for each possible data range interval of the lower bits, in this case, for example, the lower bits are 8 bits. If there is 2
This is the data correspondence that can be taken by the 8th (256) upper bits, and if the upper bits are 8 bits, a table corresponding to 256 input data may be used.

Further, the interpolation table stores the slope value of the line segment connecting the non-linear inverse quantization operation results corresponding to the individual data that the upper bits can take for each data range that the lower bits can take. This is a table in which the slope value is calculated and obtained in advance for each possible data of the high-order bit, and stored in association with the possible data of the high-order bit. Therefore, like the representative value table, this interpolation table may be a table corresponding to 256 input data if the upper bits are 8 bits.

As described above, although the present invention has three tables, each of the tables has a small table size, and even if these three tables are summed, the inverse tables corresponding to the entire range that the input data can take are obtained. Compared with the quantization tables, the data amount can be significantly reduced, the occupancy of the memory of these tables can be suppressed low, and the capacity of the memory used can be reduced.

The processing for obtaining the result of the non-linear inverse quantization operation on the compressed data when it is determined that the input data requires a high degree of accuracy is performed by using the lower bit data as a key in the inverse quantization table. , And a process of acquiring a non-linear inverse quantization operation result calculated in advance from the inverse quantization table. Therefore, by simply referring to the inverse quantization table, the appropriate nonlinear inverse quantization for the input data is performed. An operation result can be obtained.

When it is determined that the input data requires less accuracy, the process of obtaining the non-linear inverse quantization operation result on the compressed data is performed by referring to the representative value table using the upper bit data as a key. Then, the nonlinear inverse quantization operation result is obtained from this representative value table,
Further, a slope value corresponding to the upper bit is acquired from the data of the upper bit by referring to the interpolation table, and an interpolation value corresponding to a lower bit subsequent to the upper bit is obtained from the slope value. In addition to the result of the nonlinear inverse quantization operation obtained from the representative value table, to obtain the result of the nonlinear inverse quantization operation.

As described above, in the case of input data having a low degree of accuracy requirement, a simple value is obtained by referring to the representative value table, then referring to the interpolation table, and using the value obtained thereby. A nonlinear inverse quantization operation result can be obtained only by performing addition and integration.

As a specific operation, Vx is the result of the nonlinear inverse quantization operation to be obtained, V is the result of the nonlinear inverse quantization operation obtained from the representative value table, g is the slope value obtained from the interpolation table, When the lower bit data following the upper bit is X_low, Vx = V + g * X_low
It is sufficient to perform a simple operation of referring to the table twice (refer to the representative value table and the interpolation table), and performing only one integration and one addition. Moreover, the obtained result is a highly accurate and highly accurate value although it is an approximate place.
It should be noted that even in the case of input data having a low degree of accuracy, depending on the value of the input data, the result of the non-linear inverse quantization operation may be obtained only by the representative value table without performing interpolation.

[0043]

Embodiments of the present invention will be described below. The contents described in this embodiment are as follows.
The description includes the signal decoding method and the signal decoding device of the present invention, and also includes the specific processing contents of the signal decoding processing program in a recording medium on which the signal decoding processing program of the present invention is recorded.

FIG. 1 shows a configuration of a signal decoding apparatus according to the present invention. The bit stream analyzing section 1 analyzes a bit stream of compressed data as input data, and a variable length code decoding section decodes variable length encoded data. A dequantization processing unit 3 for non-linearly dequantizing the decoded data; a frequency / time conversion unit 4 for converting the non-linearly dequantized data into time axis data and outputting audio as a signal; A side information decoding unit that decodes information (referred to as side information) in which various rules necessary for the signal decoding apparatus to perform a decoding process, such as how encoding and quantization are performed, are described. 5 is provided.

As described above, the compressed data input to the signal decoding device shown in FIG. 1 converts the digital audio signal on the time axis into data on the frequency axis using orthogonal transform, Given the sensitivity characteristics of human hearing, give a large amount of information to information that is important to human hearing, and conversely, give information to information that is not so important to human hearing. This is data compressed by performing non-linear quantization, such as reducing the amount, and then performing variable-length coding on it.

Such compressed data is provided as an input to the signal decoding device of FIG. 1 and, as described above, is subjected to bit stream analysis by the bit stream analysis unit 1 and separated into side information encoded data and audio encoded data. The side information encoded data is supplied to the side information decoding unit 5
, And the encoded audio data is supplied to the variable-length code decoding unit 2.

Then, the variable-length code decoding section 2 decodes the audio coded data using the side information decoded by the side information decoding section 5 to obtain a certain length (here, 1
(Assuming 6 bits). The decoded data is subjected to a nonlinear inverse quantization process in the inverse quantization processing unit 3 to obtain, for example, data of about 32 bits (ideally, 80 bits or more).

The data which has been nonlinearly dequantized by the dequantization processing unit 3 is data on the frequency axis, which is not shown here. The data on the vertical axis is converted into data on the time axis by the frequency / time converter 4 and output as an audio signal.

The present invention has a great feature in the nonlinear inverse quantization processing performed by the inverse quantization processing unit 3, and realizes the nonlinear inverse quantization processing efficiently and with high accuracy.

Here, as the non-linear inverse quantization processing, there is a calculation term of the fourth power of is (i) in the above-mentioned equation (1). That is, as shown in FIG. 2, the data (here, 16 bits) decoded by the variable-length code decoding unit 2 is subjected to a nonlinear inverse quantization process by the inverse quantization processing unit 3, for example, 32 bits. Output as bit data,
According to the present invention, a calculation term of the fourth power of is (i) (hereinafter, is (i) is set to X and X is set to the fourth power of X) as the nonlinear inverse quantization processing is represented by the following equation. It is realized well and with high accuracy.

FIG. 3 shows an approximate curve of X to the fourth power, in which nonlinear inverse quantization of X (16-bit input data, that is, data decoded by the variable-length code decoding unit 2) is performed. It shows a curve approximating the calculation result (X to the fourth power).

As can be seen from FIG. 3, when the value of the input data X is small, the value of the fourth power of X for each input data is output as a small change with respect to the change of the input data X. However, when the value of the input data X increases, the value of the fourth power of X for each input data is output as a large change with respect to the change of the input data X.

As described above, in consideration of the sensitivity characteristics of human hearing, a large amount of information is given to information important to human hearing, and conversely, it is not very important to human hearing. This is the result of performing non-linear quantization such as reducing the amount of information given to information.

Here, if there is a table in which all the values of X that can be taken as input data are calculated to be the fourth power of X in advance, the table is set with all the values of the input data as keys. By referencing, the value of X raised to the fourth power of each input data can be immediately obtained, but when the input data is 16 bits, 2 16
As described above, it is necessary to provide a table having the result of the nonlinear inverse quantization operation corresponding to the multiplied input data, and the table size becomes enormous, which is not practical.

Therefore, according to the present invention, when input data for which a high degree of accuracy is required, that is, when the value of input data X is small, a nonlinear inverse quantization operation (X to the third power of X) is performed for each input data. Is performed in advance, and an inverse quantization table in which the operation result is stored in association with each input data is created, and the nonlinear inverse quantization operation corresponding to each input data is referred to by referring to the inverse quantization table. Ask for the result.

In the case where input data with a low accuracy requirement, that is, the value of the input data X is large, a nonlinear inverse quantization operation (X to the third power of X) is performed in correspondence with the input data at predetermined intervals. Is performed in advance, and as a representative value, the calculation result is created as a representative value, a representative value table stored in association with input data at predetermined intervals is created, and an interpolation table capable of interpolating between these representative values is created, With reference to the representative value table or both the representative value table and the interpolation table, a nonlinear inverse quantization value corresponding to each input data is obtained. Hereinafter, a description will be given using a specific example.

Here, as shown in FIG. 4A, the decoded input data X (supposed to be composed of n bits) has the rightmost side shown as the least significant bit LSB, and The left end is the most significant bit MSB. Then, in order to determine whether the value of the input data is large or small,
The input data X is divided into upper bits X_high (supposed of n1 bits) and lower bits X_low (supposed of n2 bits, and n1 + n2 = n). , It is determined whether the value of the input data is large or small.

In this embodiment, it is assumed that a boundary is set so that n1 = n2 = 8 bits for the decoded 16-bit input data X. If all the bits of the upper bit X_high are composed of “0”, that is, if the value of the upper bit X_high is 0, it is determined that the input data X has a small value and the accuracy requirement is high.
If the upper bit X_high is not 0, the input data X
Has a large value and a low accuracy requirement.

When the 16-bit input data X decoded by the variable-length code decoding unit 2 is input to the inverse quantization processing unit 3, first, the upper bit X_high and the lower bit X_low
Is divided into two by 8 bits each, and the upper bits X_h
As shown in FIG. 4B, igh and the lower bit X_low are respectively copied to registers R1 and R2 as temporary storage means. At this time, the register R1 sets the 8-bit data of the lower bit X_low to “0”, and the register R2 sets the 8-bit data of the upper bit X_high to “0”.

If the 8-bit data of the upper bit X_high is "0", it is determined that the value of the input data X is small. In this case, the inverse quantization table Q prepared in advance is referred to. Then, the result of the nonlinear inverse quantization operation (X to the fourth power) corresponding to each input data is obtained.

The inverse quantization table Q has a lower bit X
The calculation result of the fourth power of X corresponding to all the data that can be taken by _low is the content stored in association with all the data that can be taken by the lower bit X_low. Therefore, when referring to the inverse quantization table Q, the lower bits X
_low data is used as a key, and a value (result of nonlinear inverse quantization operation) corresponding to the lower bit X_low data may be obtained.

For example, if the 8-bit data of the lower bit X_low in the register R2 is "0x03" (in hexadecimal notation), when the inverse quantization table Q is referred to using "0x03" as a key, Va
For example, “0005” is obtained as the result of the nonlinear inverse quantization operation as lue.

The inverse quantization table Q has the upper bits X
When the 8 bits of _high are “0”, the result of calculating the fourth power of X corresponding to all the possible data of the 8 bits of the lower bit X_low is 8 bits of the lower bit X_low.
Since it is associated with each of the data that the bits can take, an appropriate nonlinear inverse quantization operation result can be obtained for the input data X.

As described above, when the value of the input data X is small, that is, for input data that is sensitive to human hearing and requires a high degree of accuracy, the input data X is referred to simply by referring to the inverse quantization table Q. , An appropriate nonlinear inverse quantization operation result can be obtained.

In addition, since the process at this time involves only performing one table reference (referring to the inverse quantization table Q), high-speed processing is possible. In this embodiment, as described above, the inverse quantization table Q stores the lower bits X_lo when the 8 bits of the upper bits X_high are “0”.
Since it is only necessary to have the calculation result (non-linear inverse quantization operation result) for each data that can be taken by 8 bits of w, the table size is significantly larger than the case where the calculation result corresponds to the entire input data range. It is possible to reduce the size.

In this embodiment, the input data (the data decoded by the variable length code decoding unit 2) is 16 bits.
In order to create an inverse quantization table that can handle all input data, a table having nonlinear inverse quantization operation results corresponding to as many as 2 @ 16 input data is required. Although the size is enormous, the inverse quantization table Q is 2-8
What is necessary is just to have the result of the nonlinear inverse quantization operation corresponding to the multiplied input data.

The above is the non-linear inverse quantization processing when the eight bits of the upper bit X_high are “0”, that is, when the value of the input data is small (the accuracy requirement is high). The non-linear inverse quantization processing when the eight bits of the upper bit X_high are not “0”, that is, when the value of the input data is large (the accuracy requirement is low) will be described with reference to FIG.

Also in this case, when the 16-bit input data X is input to the inverse quantization processing unit 3, first, as shown in FIG. 5A, the upper bit X_high and the lower bit X_low are converted into n1 bits and n2 bits. Divided into bits (n1 = n2 = 8),
The data of the upper bit X_high and the lower bit X_low are copied to registers R1 and R2, respectively, as temporary storage means, as shown in FIG. At this time, the register R1 sets the 8-bit data of the lower bit X_low to “0”, and the register R2 sets the 8-bit data of the upper bit X_high to “0”.

After the upper bit X_high and the lower bit X_low of the input data X are copied to the registers R1 and R2 in this way, the upper bit X_hig of the register R1 is copied.
It is determined whether the data of h is “0” or not. If it is not “0”, it is determined that the value of the input data X is large.

In FIG. 5, there are a representative value table V and an interpolation table G. The representative value table V calculates the above-mentioned fourth power of X for each data that can be taken by the high-order bit X_high, and the calculation result (nonlinear inverse quantization operation result) shows the possible result of the high-order bit X_high. 3 is a table stored in association with the data of FIG. Therefore, the upper bit X_hig in the input data
The non-linear inverse quantization operation result corresponding to the data of the upper bit X_high can be obtained using the data of h as a key.

[0071] Incidentally, the representative value table V is ² with respect to the input data X ⁿ² (This n2 is the number of bits of the lower bit X_low, since the n2 = 8 in this ^embodiment, 2 n2 = 256) The result is the calculation result of the fourth power of X corresponding to the upper bit X_high for each.

This will be described with reference to FIG. The curve shown in FIG. 6 is a graph of X of the input data X shown in FIG.
In FIG. 6, if the boundary between the upper-order bit X_high and the lower-order bit X_low is the position shown in FIG. 6, the inverse quantization table Q becomes A fourth power of X is calculated for each possible data of the lower bits X_low, and a table storing the calculation results (nonlinear inverse quantization calculation results) is obtained. In other words, if the lower bit X_low is 8 bits, a table storing the calculation result of X to the fourth power is stored for each of 2 8 (256) data.

On the other hand, the representative value table V has a range exceeding the calculation result of the fourth power of X of the inverse quantization table Q, and 2 ⁿ² (this embodiment) Since n2 = 8, 2 ⁿ² = 256)
Is a table in which the calculation result of the fourth power of X (non-linear inverse quantization operation result) corresponding to each possible data of the high-order bit X_high for each is stored as a representative value.

For example, in FIG. 6, upper bits X_h
The calculation result of X to the fourth power (non-linear inverse quantization operation result) of the data Xh of igh is Vh, the upper bit X_hig
The calculation result of the fourth power of X with respect to the data Xi of h (the result of the nonlinear inverse quantization operation) is Vi, and the data of the upper bit X_high is the calculation result of the fourth power of X with respect to Xj (the nonlinear inverse quantization operation). The result is a table such as Vj in which the calculation result of X to the fourth power (non-linear inverse quantization operation result) is stored as a representative value at every predetermined interval (in this example, 2 ⁿ² = 256).

It should be noted that the result of the non-linear inverse quantization operation for each possible data of the upper bit X_high (see FIG. 6)
In, for example, the nonlinear inverse quantization operation results Vh and V
The non-linear inverse quantization operation results (between non-linear inverse quantization operation results such as i, Vi and Vj, and Vj and Vk) are obtained using the interpolation table G.

The interpolation table G is composed of lower bits X_low
The slope value of a line segment (L1, L2, L3, etc. in FIG. 6) connecting the non-linear inverse quantization operation results corresponding to each possible data of the upper bits X_high for each possible data range (256 in this example) Is calculated in advance for each possible data of the upper bit X_high, and its slope value is stored in association with each possible data of the upper bit X_high. The method of this interpolation will be described later.

Assuming that the value of the decoded input data X (16 bits) is "0x0215" (expressed in hexadecimal notation), the upper bit X_high and the lower bit X_high X_low is divided into 8 bits each, and the 8-bit data “0x02” of the upper bit X_high is copied to the register R1, and the 8-bit data “0x15” of the lower bit X_low is copied to the register R2.

Here, since the 16-bit input data X is "0x0215", the data of the upper 8 bits of X_high is "0x02" and is not "0", so that the value of the input data X is large. to decide.

Therefore, in this case, the upper bit X_high
With reference to a representative value table V as shown in FIG. 5 using the data “0x02” as a key, the data of the upper 8 bits is represented as a representative value (result of a non-linear inverse quantization operation calculated in advance) for “0x02”. In this example, "014
9 "is obtained.

However, this “0149” is stored in the register R
This is a non-linear inverse quantization result of the 8-bit data “0x02” of the upper bit X_high stored in “1”. That is, in the 16-bit input data X, the upper 8 bits of the upper bit X_high are “0x02” and the lower 8 bits of the lower bit X_low are “0x00”. Therefore, the lower bit X_low of the actual input data X “0x0215”
It is necessary to obtain a non-linear inverse quantization operation result in consideration of the 8-bit data "0x15" (stored in the register R2). In the present invention, the lower bit X_low of 8
Interpolation is performed to obtain a result of the nonlinear inverse quantization operation in consideration of the data for the bits.

In FIG. 6 described above, each representative value (for example, Vh, Vi, Vj,...) Is represented by certain data Xh, Xi, Xj,. The result of the non-linear inverse quantization operation, which is obtained by actually calculating x to the fourth power, is shown as a representative value for each of these representative values Vh, Vi, and
Vj,... Are stored in the representative value table V in the upper bit X
_high is stored in association with each possible data.

The representative values Vh, Vi, V
are linearly interpolated by connecting j,... with straight lines L1, L2, L3,. That is, the register R
For the data of 8 bits of the upper bit X_high stored in 1, a calculation result (non-linear inverse quantization operation result) calculated in advance from the representative value table V is obtained, and the lower order stored in the register R2 is obtained. For data of 8 bits of bit X_low, its value is obtained by linear interpolation,
The value obtained by this linear interpolation is represented by a representative value table V
Is added to the calculation result calculated in advance to obtain the non-linear inverse quantization operation result to be obtained for the input data X.

As a method for obtaining a value for the lower 8 bits of data stored in the register R2 by linear interpolation, a slope value for each of the upper 8 bits of the upper bit X_high is calculated, and The gradient value is stored in the interpolation table G in association with the data of the eight bits of the upper bit X_high. As a result, the data is used as the key for each of the 8 bits of the upper bit X_high, the corresponding slope value is obtained, and the lower bit X_low is multiplied by the obtained slope value.

Here, the data of the high-order bit X_high stored in the register R 1 is used as a key, the nonlinear inverse quantization operation result (representative value) obtained from the representative value table V is set to V, and the slope obtained from the interpolation table G is set to V. If the value is represented by g and the lower bit data (data stored in the register R2) following the upper bit X_high is represented by X_low, the non-linear inverse quantization operation result Vx of the input data X to be obtained is obtained.
Can be expressed as: Vx = V + g * X_low (2)

Here, a description will be given using a specific example. As described above, the decoded 16-bit input data X is “0x
0215 ". As shown in FIG. 5, when the representative value table V is referred to using the 8-bit data “0x02” of the upper bit X_high of the input data X as a key, “014” is set as the value for the input data “0x02”.
9 "is obtained. This “0149” corresponds to V in the equation (2).

Next, the upper bit X of the input data X
When the interpolation table G is referred to using the 8-bit data “0x02” of _high as a key, “0x02” is represented by “0”.
003 ”is obtained. This "0003" is (2)
It corresponds to g in the formula. These values are expressed as V,
g, and 8 bits of lower bits X_low are substituted for X_low in the equation (2), and the equation (2) is calculated. An approximated value can be obtained as a result of the nonlinear inverse quantization operation.

This will be described with reference to FIG. "0x021
Assuming that the input data X of "5" is a value existing between Xi and Xj in FIG. 6, the inverse quantization operation result to be obtained at that time is Vi + Va.

That is, the input data X of “0x0215”
8 bits of data "0x0"
2 corresponds to Xi, and the data “0x0”
Referring to the representative value table V with “2” as a key, FIG.
As shown in “1”, “0149” is acquired as Vi. Referring to the interpolation table G using the data “0x02” as a key, as shown in FIG.
“0003” is acquired as the slope value g at that time.

Then, by using the lower bit X_low stored in the register R2, the data of the lower bit X_low is multiplied by the slope value g to obtain Va. Then, by adding this Va and the representative value Vi obtained from the representative value table V, the result of the nonlinear inverse quantization operation for the input data X of “0x0215” can be obtained.

The upper bit X_high of the input data X
, Xh, Xi, Xj,... Shown in FIG. 6, the non-linear inverse quantization operation result for the input data can be directly obtained from the representative value table V. Therefore, it is not necessary to perform interpolation. Absent.

FIG. 7 is a flowchart showing the above-described processing procedure, which is a generalized description of the processing procedure of the present invention. In FIG. 7, input data X (data decoded by the variable length code decoding unit) is first divided into upper bits X_high and lower bits X_low (step s1), and the upper bits X_high are “0”. It is checked whether or not the upper bit X_high is “0” (step s2). If the lower bit X_low is a key, the previously generated inverse quantization table Q is referenced (step s3).

As described above, the upper bit X_high is "0".
In the case of, the nonlinear inverse quantization operation result (X to the third power) to be obtained can be obtained only by referring to the inverse quantization table Q.

On the other hand, if it is determined in step s2 that the upper bit X_high is not “0”, the representative bit table V is referred to using the upper bit X_high as a key,
The representative value V for the upper bit X_high is obtained, and the gradient value g at that time is obtained by referring to the interpolation table G using the upper bit X_high as a key (step s4).

Then, the representative value V obtained in step s4
, The slope value g, and the lower bits X_l divided in step s1
Using ow, the above equation (2), that is, V + g * X_low
Is calculated (step s5), and the input data X at that time is calculated.
To obtain the result of the nonlinear inverse quantization operation.

By performing the above-mentioned processing, the amount of calculation of the nonlinear inverse quantization processing to be performed by the inverse quantization processing unit 3 can be significantly reduced. That is, the processing to be performed by the inverse quantization processing unit 3 includes the variable-length code decoding unit 2
There is a calculation of the fourth power of the input data X decoded in (4) (see FIG. 2), but in the present invention, the calculation amount of this part can be greatly reduced.

In order to perform this processing, as described above, for each of the input data that can be taken by the decoded input data X (2 16 powers if the input data is 16 bits), the aforementioned X It is conceivable to perform the calculation of the fourth power of each time, or to have a table in which the result of the nonlinear inverse quantization operation corresponding to each input data is described for each input data. And the table size becomes extremely large.

On the other hand, according to the present invention, the input data X is divided into an upper bit X_high and a lower bit X_low, and when the upper bit X_high is 0, the value of the input data X is small (accuracy is required). ), The lower bit X_low of the input data X at that time is used as a key to refer to the inverse quantization table Q to obtain a value of X to the third power.

This inverse quantization table Q is obtained by input data X
Is not created corresponding to the entire range that can be taken, but only needs to correspond to the data range that the lower bit X_low can take, so that the table size can be significantly reduced.

If the upper bit X_high is not 0, it is determined that the value of the input data is large (accuracy is not so much required). In this case, the representative bit is set using the upper bit X_high of the input data X at that time as a key. By referring to the table V, a representative value V for the upper bit X_high is obtained, and the gradient value g at that time is obtained by referring to the interpolation table G using the upper bit X_high as a key.

Then, using the representative value V, the slope value g, and the lower-order bit X_low at that time, V +
By calculating g * X_low, a value obtained by approximating the fourth power of X to the input data X at that time can be obtained as a result of the nonlinear inverse quantization operation.

As described above, the representative value table V at this time calculates the fourth power of X for each 2 ⁿ² (n2 is the number of lower bits X_low) for the input data X. A table having the result (nonlinear inverse quantization operation result) as a representative value is obtained. For example, the lower bit X_low
Is 8 bits (n2 = 8), the result is a table in which the results of the non-linear inverse quantization operation every 2 to the eighth power (= 256) are stored as representative values.

Then, a slope value g is calculated in advance for each representative value, and the slope value g and the upper bit X are calculated.
An interpolation table G has a correspondence between each data that can be taken by _high. Even when all of the representative value table V, the interpolation table G, and the inverse quantization table Q corresponding to the above-mentioned lower bits X_low are combined, the total data amount corresponds to the entire data range that the input data X can take. The data amount can be much smaller than the data amount of the table.

By the way, if the input data is 16 bits, the entire data range that the input data can take is 2 to the 16th power, and more than 60,000 input data
However, in the present invention, even if all the tables are summed up, the data amount can be much smaller than that, and a small-capacity memory can be used. In the example described above, each of the representative value table V, the interpolation table G, and the inverse quantization table Q can be a table corresponding to 2 8 (= 256) data, and each of the tables is a small table. Can be size.

Also, as a processing procedure, upper bits X_h
When igh is 0, a nonlinear inverse quantization operation result can be obtained only by referring to the inverse quantization table once,
Even when the upper bit X_high is not 0, the nonlinear inverse is performed only by performing simple integration and addition as shown in Expression (2) after referring to the representative value table V and the interpolation table G once each. Since the result of the quantization operation can be obtained, even an inexpensive CPU with a not so high processing capability can cope sufficiently.

The present invention is not limited to the embodiment described above, but can be variously modified without departing from the gist of the present invention. For example, in the above-described embodiment, an audio signal has been described as an example of a signal to be processed. However, the present invention can be applied not only to an audio signal but also to any signal requiring processing as described in the present invention. . Also, specific numerical values (for example, the number of bits of decoded input data, the number of bits of upper bits X_high,
Of course, the number of bits of the lower bit X_low) is merely an example, and the present invention is not limited to this example.

Further, according to the present invention, a processing program in which a processing procedure for realizing the present invention described above is described is created, and the processing program is recorded on a recording medium such as a floppy disk, an optical disk, or a hard disk. The present invention also includes a recording medium on which the processing program is recorded. Further, the processing program may be obtained from a network.

[0107]

As described above, according to the present invention, the decoded input data is divided into upper bits and lower bits, and when the upper bits are 0, the value of the input data is small ( (Accuracy is required), and in that case, the lower-order bit data of the input data at that time is used as a key to refer to the inverse quantization table to obtain a previously calculated nonlinear inverse quantization operation result. This inverse quantization table is not created in correspondence with the entire range of the input data, but only needs to correspond to the data range of the lower bits, so that the table size can be significantly reduced. .

If the upper bit is not 0, it is determined that the value of the input data is large (it does not require much accuracy). In this case, the upper bit of the input data at that time is used as a key to refer to the representative value table. To obtain the representative value for the upper bits, and obtain the gradient value at that time by referring to the interpolation table as the upper bits. Then, using the representative value, the slope value, and the lower-order bit at that time, the result of the nonlinear inverse quantization operation can be obtained for the input data at that time by a simple calculation. In addition,
At this time, the representative value table indicates that 2
^It is a table for each ⁿ² (n2 is the number of lower bits). Even if all of the representative value table, the interpolation table, and the inverse quantization table corresponding to the lower bits are combined, the total data amount is larger than the data amount of the table corresponding to the entire data range that the input data can take. Can have a much smaller data amount.

By the way, if the input data is 16 bits, the entire data range that the input data can take is 2 to the 16th power, and more than 60,000 input data are
However, in the present invention, even if all the tables are summed up, the table amount can be significantly smaller than that, and a small-capacity memory can be used.

In the processing procedure, the upper bit is set to 0.
, The nonlinear inverse quantization operation result can be obtained only by referring to the inverse quantization table once. Even when the upper bit is not 0, the representative value table and the interpolation table are each read once. After the reference, the result of the nonlinear inverse quantization operation can be obtained only by performing simple integration and addition. Therefore, even an inexpensive CPU with a not so high processing capability can be sufficiently used.

As described above, according to the present invention, non-linear inverse quantization processing when decoding data that has been transformed into the frequency domain and then quantized, coded and compressed can be efficiently and accurately performed with a small amount of computation. It can be carried out.

In particular, the present invention exerts its effects by being applied to a device that reproduces an audio signal that has been converted into such a frequency domain, quantized, coded, and compressed, and the effects of the present invention are exhibited. In addition, the amount of calculation of the nonlinear inverse quantization process performed in the decoding process can be significantly reduced, and the amount of data such as tables required for the calculation can be significantly reduced. As a result, a cheap CPU and memory can be used, and the price and the size and weight of the device itself can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of a signal decoding device according to the present invention.

FIG. 2 is a diagram illustrating an outline of a nonlinear inverse quantization operation performed by an inverse quantization processor illustrated in FIG. 1;

3 is a diagram showing a curve approximating the result of the nonlinear inverse quantization operation (fourth power of input data X) shown in FIG. 2;

FIG. 4 is a diagram illustrating a processing procedure for obtaining a result of a nonlinear inverse quantization operation when input data is small (when accuracy is required to be high) in the signal decoding process of the present invention.

FIG. 5 is a diagram illustrating a processing procedure for obtaining a result of a nonlinear inverse quantization operation when input data is large (accuracy is low required) in the signal decoding processing of the present invention.

FIG. 6 is a diagram illustrating a case where input data is large (accuracy is low required), using a curve approximating the result of the nonlinear inverse quantization operation (input data X raised to the fourth power) shown in FIG. 3; FIG. 9 is a diagram illustrating a process of obtaining a nonlinear inverse quantization operation result by referring to a value table and an interpolation table.

FIG. 7 is a flowchart illustrating an overall processing procedure of the present invention.

[Explanation of symbols]

 Reference Signs List 1 bit stream analysis unit 2 variable length code decoding unit 3 inverse quantization processing unit 4 frequency / time conversion unit 5 side information decoding unit X_high upper bit X_low lower bit R1 Register for storing data of upper bit X_high R2 lower bit X_low Register for storing the data of the above data Q Inverse quantization table V Representative value table (acquired representative values) G Interpolation table g Slope value X Input data

Claims (15)

[Claims] 1. After being converted into data on a frequency axis,
Non-linear quantization according to the required degree of accuracy is performed, and compressed data compressed by being encoded is used as input data, and after decoding the compressed data as the input data, the compressed data corresponding to the required degree of accuracy is determined. A signal decoding method for performing non-linear inverse quantization processing and converting it into data on a time axis, wherein for the input data having a high degree of accuracy requirement, each of the decoded input data is used as a key, and Prepare an inverse quantization table that can obtain the result of the nonlinear inverse quantization operation corresponding to the compressed data. For the input data with a low degree of required accuracy, each input data decoded at predetermined intervals is used as a key. A representative value table that can be used as a representative value to obtain the result of the nonlinear inverse quantization operation corresponding to the input data at the predetermined interval is prepared, and interpolation is performed between these representative values. When performing a non-linear inverse quantization process according to the required degree of accuracy, a required degree of accuracy of the decoded input data is determined, and an input having a high required degree of accuracy is performed. If it is determined that the input data is a data, the input data is referred to the dequantization table as a key to obtain a non-linear inverse quantization operation result for the input data. When it is determined that the input data is the key, the representative value table or both the representative value table and the interpolation table are referred to as the key to obtain a nonlinear inverse quantization operation result for the input data. Characteristic signal decoding method. 2. The method according to claim 1, wherein the degree of accuracy of the input data is determined by setting the number of bits of the decoded input data to n (n is a positive integer) and setting the input data to upper bits (the number of bits is n1). ) And lower bits (the number of bits is n2, n1 + n2 = n), and it is determined whether the data of the upper bits is 0. If the data of the upper bits is 0, The demand for accuracy of input data is high,
2. The signal decoding method according to claim 1, wherein when the data of the upper bit is not 0, it is determined that the required degree of accuracy of the input data is low. 3. The inverse quantization table according to claim 1, wherein a non-linear inverse quantization operation result calculated in advance for each data in the data range of the lower bits is associated with each data of the lower bits. The representative value table stores the non-linear inverse quantization operation result calculated in advance for each possible data of the upper bits for each possible data range interval of the lower bits. A table stored in association with each possible data of a high-order bit, wherein the interpolation table is a non-linear inverse of each possible data of the high-order bits for each possible data range of the low-order bits. The slope value of the line connecting each quantization operation result is reserved for each possible data of the upper bit side. Calculated advance determined, characterized in that the slope value is a table which is stored in association with each of the data may take the upper bits claim 2
The signal decoding method according to the above. 4. A process for obtaining a result of a non-linear inverse quantization operation on the input data when it is determined that the input data requires a high degree of accuracy, includes converting data represented by lower bits of the input data into a key. Refers to the inverse quantization table as a process of obtaining a non-linear inverse quantization operation result calculated in advance from the inverse quantization table, in the case where it is determined that the input data of the low required degree of accuracy The process of obtaining the non-linear inverse quantization operation result for the input data refers to the representative value table by using data represented by the upper bits of the input data as a key, and performs nonlinear inverse quantization calculated in advance from the representative value table. When the calculation result is obtained and interpolation corresponding to the data of the lower bit following the upper bit is necessary, the interpolation is performed in advance from the representative value table. After obtaining the calculated non-linear inverse quantization operation result, the upper bit data is used as a key to obtain the gradient value corresponding to the upper bit by referring to the interpolation table, and from the gradient value to the higher bit. An interpolated value corresponding to the data of the following lower bits is obtained, and the interpolated value is obtained as a non-linear dequantized operation result by adding the interpolated value to the non-linear dequantized operation result obtained from the representative value table. 3. The signal decoding method according to 3. 5. A non-linear inverse quantization operation result for the input data when it is determined that the input data has a low degree of required accuracy is Vx, a non-linear inverse quantization operation result to be obtained from the representative value table. When the obtained nonlinear inverse quantization operation result is represented by V, the gradient value obtained from the interpolation table is represented by g, and the data of the lower bit following the upper bit is represented by X_low, Vx = V + g * X_low is obtained. The signal decoding method according to claim 4, wherein 6. After being converted into data on the frequency axis,
Non-linear quantization according to the required degree of accuracy is performed, and compressed data compressed by being encoded is used as input data, and after decoding the compressed data as the input data, the compressed data corresponding to the required degree of accuracy is determined. What is claimed is: 1. A signal decoding device for performing a non-linear inverse quantization process and converting it into data on a time axis, said signal decoding device comprising: a decoding unit for decoding said encoded input data; Inverting means for nonlinearly quantizing the data and frequency / time converting means for converting the non-linearly inversely quantized data into data on the time axis are included as constituent elements. Is high as the table for the input data, the inverse quantization that can obtain the nonlinear inverse quantization operation result corresponding to each input data using the decoded input data as a key As a table for the input data having a low degree of accuracy requirement, the decoded input data at each predetermined interval is used as a key to represent the result of the nonlinear inverse quantization operation corresponding to the input data at the predetermined interval. It has a representative value table that can be obtained as a value, and further has an interpolation table that can interpolate between these respective representative values. When performing a nonlinear inverse quantization process according to the required degree of accuracy, decoding is performed. The degree of accuracy required of the input data is determined, and when it is determined that the input data has a high degree of accuracy required, the input data is referred to by referring to the inverse quantization table as a key. When the result of the nonlinear inverse quantization operation is obtained, and it is determined that the input data has a low degree of accuracy, the representative value table or the input value is used as the key. A signal decoding device which obtains a result of a non-linear inverse quantization operation on the input data by referring to both the representative value table and the interpolation table. 7. The method according to claim 6, wherein the degree of accuracy of the input data is determined by setting the number of bits of the decoded input data to n (n is a positive integer) and setting the input data to upper bits (the number of bits is n1). ) And lower bits (the number of bits is n2, n1 + n2 = n), and it is determined whether the data of the upper bits is 0. If the data of the upper bits is 0, The demand for accuracy of input data is high,
7. The signal decoding apparatus according to claim 6, wherein when the data of the upper bit is not 0, it is determined that the required degree of accuracy of the input data is low. 8. The inverse quantization table associates a non-linear inverse quantization operation result calculated in advance for each data in the data range that the lower bits can take with each possible data of the lower bits. The representative value table stores the non-linear inverse quantization operation result calculated in advance for each possible data of the upper bits for each possible data range interval of the lower bits. A table stored in association with each possible data of a high-order bit, wherein the interpolation table is a non-linear inverse of each possible data of the high-order bits for each possible data range of the low-order bits. The slope value of the line connecting each quantization operation result is reserved for each possible data of the upper bit side. Calculated advance determined, claim 7 in which the inclination value is characterized in that it is a table stored in association with each of the data may take the upper bits
The signal decoding device according to any one of the preceding claims. 9. A process for obtaining a result of a non-linear inverse quantization operation on the input data when it is determined that the input data requires a high degree of accuracy, includes converting data represented by lower bits of the input data into a key. Refers to the inverse quantization table as a process of obtaining a non-linear inverse quantization operation result calculated in advance from the inverse quantization table, in the case where it is determined that the input data of the low required degree of accuracy The process of obtaining the non-linear inverse quantization operation result for the input data refers to the representative value table by using data represented by the upper bits of the input data as a key, and performs nonlinear inverse quantization calculated in advance from the representative value table. When the calculation result is obtained and interpolation corresponding to the data of the lower bits following the upper bits is necessary, the interpolation is performed in advance from the representative value table. After obtaining the calculated non-linear inverse quantization operation result, the upper bit data is used as a key to obtain the gradient value corresponding to the upper bit by referring to the interpolation table, and from the gradient value to the higher bit. An interpolated value corresponding to the data of the following lower bits is obtained, and the interpolated value is obtained as a non-linear dequantized operation result by adding the interpolated value to the non-linear dequantized operation result obtained from the representative value table. 9. The signal decoding device according to 8. 10. The result of the nonlinear inverse quantization operation on the input data when it is determined that the input data has a low degree of accuracy requirement is Vx, the result of the nonlinear inverse quantization operation to be obtained from the representative value table. When the obtained nonlinear inverse quantization operation result is represented by V, the gradient value obtained from the interpolation table is represented by g, and the data of the lower bit following the upper bit is represented by X_low, Vx = V + g * X_low is obtained. The signal decoding device according to claim 9, wherein 11. Compressed data that has been converted into data on the frequency axis, subjected to non-linear quantization according to the degree of accuracy required, and further encoded to be compressed data is used as input data. After decoding the compressed data of the non-linear inverse quantization processing according to the required degree of accuracy, a recording medium that stores a signal decoding processing program that converts it to data on the time axis, the processing program, When performing the non-linear inverse quantization process according to the required accuracy, a procedure for determining the required accuracy of the decoded input data, and a case where it is determined that the input data has a high required accuracy The input data is used as a key to refer to the inverse quantization table that can obtain the result of the nonlinear inverse quantization operation for each input data. The procedure for obtaining the calculation result, and when it is determined that the compressed data has a low accuracy requirement, the input data corresponding to the predetermined interval is used as a key using the input data at each predetermined interval as a key. A non-linear inverse quantization operation result for the input data is obtained by referring to a representative value table that can acquire the result of the non-linear inverse quantization operation as a representative value or both an interpolation table that can interpolate between the representative value table and each representative value. A recording medium recording a signal decoding processing program, comprising: 12. Determining the required degree of accuracy of the input data, when the number of bits of the decoded input data is n (n is a positive integer), the input data is set to upper bits (the number of bits is n1 ) And lower bits (the number of bits is n2, n1 + n2 = n), and it is determined whether the data of the upper bits is 0. If the data of the upper bits is 0, Demand for accuracy of input data is high,
12. The recording medium according to claim 11, wherein when the data of the upper bit is not 0, it is determined that the required degree of accuracy of the input data is low. 13. The inverse quantization table associates a non-linear inverse quantization operation result calculated in advance for each data in the data range that the lower bits can take with each possible data of the lower bits. The representative value table stores the non-linear inverse quantization operation result calculated in advance for each possible data of the upper bits for each possible data range interval of the lower bits. A table stored in association with each possible data of a high-order bit, wherein the interpolation table is a non-linear inverse of each possible data of the high-order bits for each possible data range of the low-order bits. The slope value of the line connecting each quantization operation result is calculated according to each possible data of the upper bit side. Advance determined by fit calculation, claim 1 in which the gradient value is characterized in that it is a table stored in association with each of the data may take the upper bits
A recording medium on which the signal decoding program according to Item 2 is recorded. 14. A process for obtaining a result of a non-linear inverse quantization operation on the input data when it is determined that the input data requires a high degree of accuracy, includes converting data represented by lower bits of the input data into a key. Refers to the inverse quantization table as a process of obtaining a non-linear inverse quantization operation result calculated in advance from the inverse quantization table, in the case where it is determined that the input data of the low required degree of accuracy The process of obtaining the non-linear inverse quantization operation result for the input data refers to the representative value table by using data represented by the upper bits of the input data as a key, and performs nonlinear inverse quantization calculated in advance from the representative value table. When the calculation result is obtained and interpolation corresponding to the data of the lower bit following the upper bit is required, the interpolation is performed in advance from the representative value table. After obtaining the calculated non-linear inverse quantization operation result, the upper bit data is used as a key to obtain the slope value corresponding to the upper bit by referring to the interpolation table, and from the slope value to the upper bit. An interpolated value corresponding to the data of the following lower bits is obtained, and the interpolated value is obtained as a non-linear dequantized operation result by adding the interpolated value to the non-linear dequantized operation result obtained from the representative value table. 14. A recording medium on which the signal decoding program according to 13 is recorded. 15. The result of the nonlinear inverse quantization operation on the input data when it is determined that the input data has a low degree of accuracy requirement is Vx, the result of the nonlinear inverse quantization operation to be obtained from the representative value table. When the obtained nonlinear inverse quantization operation result is represented by V, the gradient value obtained from the interpolation table is represented by g, and the data of the lower bit following the upper bit is represented by X_low, Vx = V + g * X_low is obtained. A recording medium on which the signal decoding processing program according to claim 14 is recorded.