JPH08162966A - Audio data coder and decoder - Google Patents

Abstract

PURPOSE: To reduce quantization noise generated in accordance with the coding and decoding of audio data. CONSTITUTION: This device is provided with an unequal space quantization coding circuit 9 dividing a value area that mantissa part data is possible to take into plural areas at an unequal space according to the probability distribution that mantissa part data is actually generated, assigning each code to each of plural divided sections and performing the quantization coding of mantissa part data. By performing the quantization coding of audio data according to the probability distribution that mantissa part data is actually generated, an inconvenience that many codes are assigned to data with small generation probability can be eliminated and the generation of a quantization error can be reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio data coding apparatus and a decoding apparatus, and is particularly suitable for use in an apparatus for highly efficient coding and decompressing audio data.

[0002]

2. Description of the Related Art Conventionally, techniques for highly efficient encoding of audio data include band division encoding in which audio data is divided into a plurality of bands on the time axis and encoded, or audio data is orthogonalized on the frequency axis. There is transform coding that transforms and divides into a plurality of bands for coding. There is also high-efficiency coding in which these codings are combined to divide audio data into a plurality of bands on the time axis, and each band signal is further orthogonally converted and coded on the frequency axis.

As an example of the conventional technique, transform coding and decoding using MDCT (Modified Discrete Cosine Transform) as shown in FIG. 3 will be described here. Regarding the contents of MDCT, see Japanese Patent Laid-Open No. 5-18344.
The details are described in the conventional example of Japanese Patent Publication No.

First, the encoding of audio data will be described. In the encoding device shown in FIG. 3A, the audio data input from the audio data input terminal 1 is subjected to windowing by the windowing circuit 2 in block units, for example, 512 samples as one block. That is, in the windowing circuit 2, 50% of the data of the current block is overlap-added to the data of the previous block.
As a result, the connection between blocks at the time of decoding is made smooth.

The result of the windowing process by the windowing circuit 2 is input to the MDCT circuit 3. In the MDCT circuit 3, orthogonal transform is performed by MDCT (Modified Discrete Cosine Transform), and the resulting MDCT coefficient data is input to the block floating point transform circuit 4. As described below, the MDCT coefficient data is coded by dividing it into an exponent part and a mantissa part.

That is, the block floating point conversion circuit 4
Then, 512 MDCs input from the MDCT circuit 3
Of the T coefficient data, for example, 240 MDCT coefficient data from the zero frequency are divided into several bands (for example, 64 bands), and a block floating point of one exponent part data and a plurality of mantissa part data per one band. Converted to data. At this time, the number of coefficient data in one band is, for example, 1 to 8, and the band division method is adopted in which the number of coefficient data in one band increases from the low frequency region to the high frequency region. .

Then, the exponent part data of the block floating point data converted as described above is input to the bit allocation circuit 5, and the mantissa part data is input to the equal interval quantization coding circuit 6. In the bit allocation circuit 5, the quantization bit length of the mantissa data is determined based on the exponent data input from the block floating point conversion circuit 4 by using auditory masking, and the determined quantization bit length is determined. The length is input to the equally-spaced quantization coding circuit 6.

In the uniform interval quantization coding circuit 6, M of the mantissa part data input from the block floating point conversion circuit 4 is input.
Data of the quantized bit length inputted from the bit allocation circuit 5 is taken out from SB (most significant bit), and the quantized coding process is performed by the taken out data. Then, the data output terminals 7 and 8 output the exponent part data and the quantized coded data of the mantissa part data, which are recorded on a recording medium (not shown) or transmitted via a transmission path.

Next, decoding of audio data will be described. In the decoding device shown in FIG. 3B, the exponent part data and the quantized coded data of the mantissa part data recorded on a recording medium (not shown) or transmitted on a transmission path are input to the data input terminals 10 and 11, respectively. Are input to the bit allocation circuit 12 and the equally-spaced inverse quantization / decoding circuit 13 via.

In the bit allocation circuit 12, the quantization bit length of the mantissa part data is determined based on the exponent part data input from the data input terminal 10 by using auditory masking, as in the case of encoding. Then, the determined quantized bit length is input to the equal interval inverse quantization decoding circuit 13.

In the equal interval inverse quantization decoding circuit 13, data is input from the data input terminal 11 by the amount of the quantization bit length input from the bit allocation circuit 12, and the input data is inverse quantization decoded. Then, this is input to the block floating point inverse conversion circuit 14 as mantissa data.

In the block floating point inverse conversion circuit 14,
The exponent part data input from the data input terminal 10 and the mantissa part data input from the equal-interval inverse quantization / decoding circuit 13 are converted into floating-point or fixed-point coefficient data, which is IMDCT (Modified Discrete Cosine Inverse Transform). ) Is input to the circuit 15.

In the IMDCT circuit 15, the coefficient data input from the block floating point inverse transform circuit 14 is subjected to inverse MDCT processing in block units, for example, 512 samples as one block, and the processing result is subjected to the windowing circuit 16 Entered in. In the windowing circuit 16, windowing processing is performed for each block, and 50% of the data of the current block is overlap-added to the data of the previous block. The audio data thus obtained is output from the audio data output terminal 17.

[0014]

By the way, the quantization of the mantissa data of the MDCT coefficient data, which is performed in the equal-interval quantization encoding circuit 6, is performed by dividing -1 to +1 at equal intervals. It is a uniform quantization. However, in general, mantissa data does not occur evenly between -1 and +1. Therefore, there is a problem that equal-quantized quantization causes a large amount of quantization noise.

In particular, when the number of coefficient data in one band is one, the mantissa data is generated between -1 and -0.5 and between +0.5 and +1. 0.5 to +
It does not occur between 0.5. Nevertheless, in equal-spaced quantization, since codes are evenly assigned even between −0.5 and +0.5, there is a problem that quantization noise becomes large.

The present invention has been made to solve such a problem, and an object of the present invention is to provide an audio data coding apparatus and a decoding apparatus capable of reducing quantization noise. And

[0017]

An audio data coding apparatus according to the present invention performs an orthogonal transform on audio data on a frequency axis, and coefficient data obtained by this coding is divided into exponent part data and mantissa part data for coding. In the audio data encoding device, the range of values that the mantissa data can take is divided into a plurality of sections at unequal intervals according to the probability distribution in which the mantissa data actually occurs, and each of the sections is divided into a plurality of sections. To each of which a code is assigned to each of the audio data and the unequal interval quantization coding means for performing the quantization coding of the audio data.

Another feature of the present invention is to obtain coefficient data by orthogonally transforming the windowing means for windowing the input audio data and the data windowed by the windowing means. Orthogonal transforming means, block floating point transforming means for block floating point transforming the coefficient data obtained by the orthogonal transforming means to obtain exponent part data and mantissa part data for each frequency band, and the block floating point transforming means. The bit allocation means comprises: a bit allocation means for allocating a quantized bit length of the mantissa part data based on the exponent part data obtained by; and a table determined according to a probability distribution in which the mantissa part data occurs. Information of the quantization bit length of the mantissa data and the information of the frequency band to which the mantissa data belongs In searching for the table are those having a non-equidistant quantization coding means for performing quantization encoding of the mantissa data.

Another feature of the present invention is that the table stores the mantissa in each of the sections obtained by dividing the value range of the mantissa data into a plurality of sections so that the occurrence probabilities of the mantissa data are equal. It is characterized in that it is determined by assigning a plurality of codes respectively determined according to the quantization bit length of the partial data.

Another feature of the present invention is that in the table, all zero bits are assigned to a section in which the mantissa data generation probability is zero.

Further, the audio data decoding device of the present invention is an audio data decoding device for decoding the audio data obtained by dividing the coefficient data obtained by the orthogonal transformation into exponent part data and mantissa part data and decoding the coded audio data. In addition, it has an unequal interval inverse quantization decoding means for performing inverse quantization decoding of audio data encoded according to the probability distribution in which the mantissa data actually occurs.

Another feature of the present invention is that it is determined according to the bit allocation means for allocating the quantization bit length of the mantissa data based on the input exponent data and the probability distribution of the mantissa data. A quantization table, the table is searched based on the information of the quantization bit length of the mantissa data assigned by the bit assigning means and the information of the frequency band to which the mantissa data belongs, and the quantization code Unequal-interval inverse-quantization decoding means for inverse-quantizing-decoding the converted mantissa data, and mantissa data and exponent-part data inverse-quantized and decoded by the unequal-interval inverse-quantization decoding means And a block floating point inverse transforming means for transforming the coefficient data into coefficient data, and an inverse orthogonal transforming means for inverse orthogonal transforming the coefficient data transformed by the block floating point inverse transforming means. When those having a windowing unit for performing windowing in the inverse orthogonal transformed data by the inverse orthogonal transform circuit.

Another feature of the present invention is that, in the table, the mantissa is divided into a plurality of sections in which the value range of the mantissa data is divided so that the occurrence probabilities of the mantissa data are equal. It is characterized in that it is determined by assigning a plurality of codes respectively determined according to the quantization bit length of the partial data.

Another feature of the present invention is that in the table, a code of all bits of zero is assigned to a section where the probability of occurrence of the mantissa data is zero.

[0025]

Since the audio data encoding device of the present invention comprises the above technical means, the audio data is quantized and encoded according to the probability distribution of the values actually generated in the range of the mantissa data. Like For example, according to the invention described in claim 2, based on the information of the quantization bit length of the mantissa part data and the information of the frequency band to which the mantissa part data belongs, according to the actual occurrence probability distribution of the mantissa part data. The determined table is referred to, and the mantissa data is quantized and coded according to this table. Therefore, a large number of codes are not assigned to the data having a small occurrence probability, and the occurrence of quantization error is reduced.

Further, according to the audio data decoding apparatus of the present invention, the inverse quantization decoding of the audio data is performed according to the probability distribution of the values actually generated in the range of the mantissa data. become. For example, according to the invention described in claim 6, based on the information of the quantization bit length of the mantissa data and the information of the frequency band to which the mantissa data belongs, according to the actual occurrence probability distribution of the mantissa data. The determined table is referred to, and the quantized and encoded mantissa data is dequantized and decoded according to this table, and the occurrence of a quantization error is prevented.

[0027]

EXAMPLE An example of the present invention will be described below. In the present embodiment, as will be described below, by subdividing the range from −1 to +1 which is the range of the mantissa part data into unequal intervals according to the probability distribution in which the mantissa part data actually occurs, the coefficient Creating a table to quantize the data.

The probability distribution that the mantissa data is generated is
Since it changes according to the number of coefficient data in one frequency band, according to the present embodiment, it is not necessary to allocate a large number of codes to data having a small occurrence probability. This allows
The mantissa data can be brought closer to the data actually generated in that band, and the noise due to the quantization error can be reduced. Therefore, it is possible to improve the audible sound quality.

For example, when the quantization bit length of the mantissa data is 2 bits, if "11" is used as the inhibition code, "0" is set.
There are three possible codes, 1 ”,“ 00 ”, and“ 10. ”In this case, if the range of −1 to +1 of the mantissa data is divided evenly, the code is in the range of −1 <k <−0.333. “01” is assigned, and the average value of this range is −0.66.
7 is the mantissa data after decoding.

Similarly, -0.333 <k <+0.333
A code "00" is assigned to the range of, and the average value 0 of this range is used as the mantissa data after decoding. Further, the code “10” is assigned to the range of +0.333 <k <+1, and the average value +0.667 of this range is used as the decoded mantissa data.

However, in practice, -0.5 to +0.5
The probability of occurrence of mantissa data in the range is 0, and the average value on the negative side is -0.75, which is the average value in the range from -1 to -0.5. Therefore, if -0.667 is used as the mantissa data and is multiplied by the exponent part data to decode, then -0.75 is used as the mantissa part data and is decrypted by multiplying the exponent part data. Since it is biased, a quantization error occurs.

On the other hand, in the present embodiment, in accordance with the actual occurrence probability of the mantissa data, the range of possible values of the mantissa data is subdivided into sections so that the occurrence probabilities are equal. Since the code is assigned and quantized and encoded, it is possible to prevent the above-described quantization error from occurring.

The above will be described with reference to the drawings.
First, the encoding of audio data will be described. FIG.
In the encoding device shown in (a), the audio data input from the audio data input terminal 1 is windowed by the windowing circuit 2 in block units, for example, 512 samples as one block. That is, in the windowing circuit 2, 50% of the data of the current block is overlap-added to the data of the previous block. This allows
The connection between blocks at the time of decoding is made smooth.

The result of the windowing processing by the windowing circuit 2 is input to the MDCT circuit 3. In the MDCT circuit 3, orthogonal transform is performed by MDCT (Modified Discrete Cosine Transform), and the resulting MDCT coefficient data is input to the block floating point transform circuit 4. As described below, the MDCT coefficient data is coded by dividing it into an exponent part and a mantissa part.

That is, the block floating point conversion circuit 4
Then, 512 MDCs input from the MDCT circuit 3
Of the T coefficient data, for example, 240 MDCT coefficient data from the zero frequency are divided into several bands (for example, 64 bands), and a block floating point of one exponent part data and a plurality of mantissa part data per one band. Converted to data.

The block floating point conversion will now be described. In the block floating point conversion, first, the maximum value is found from the absolute values of the coefficient data within one band. Next, the maximum value data is represented as floating point data F by the following equation.

F = M × 2exp (−N) where F: floating point data M: mantissa data, 0.5 ≦ M <1, −1 ≦ M <-0.
5 N: exponent data, positive integer

Finally, other coefficient data within one band is divided by the exponent part 2exp (-N) of the maximum value data (floating point data) F, and the division result thereof is used as mantissa part data. Through the above processing, one exponent part data in one band and mantissa part data of the number equal to the number of coefficient data in one band are calculated.

At this time, the number of coefficient data in one band is, for example, 1 to 8, and the number of coefficient data in one band increases from the low frequency region to the high frequency region. Is taking. For example, from low frequency to high frequency, 16 bands of 1 coefficient data in 1 band are 16 bands of 1 coefficient data in 1 band.
That is, 16 bands each having 4 pieces of coefficient data in one band and 16 bands having 8 pieces of coefficient data in one band. In this case, there are 64 exponent part data and 240 mantissa part data as a whole.

FIG. 2 shows the above-mentioned band division.
Since the human auditory characteristics are dull in the high range, the band division is performed in such a manner that the data represented by one exponent part is reduced in the low range where the auditory characteristics are sharp, and one exponent part is included in the higher range. This is to represent a large number of data.

Since the 16 pieces of coefficient data from the lowest frequency are represented by one exponent part data and one mantissa part data, respectively, there is no error from the actual data.
On the other hand, in the 16 bands on the high frequency side, the eight coefficient data in each band are represented by one exponent part data and eight mantissa part data, so that a conversion error occurs. . However, since this conversion error is an error that cannot be recognized by humans, there is no problem in practical use.

Next, the exponent part data of the block floating point data converted as described above is input to the bit allocation circuit 5, and the mantissa part data is input to the unequal interval quantization encoding circuit 9. In the bit allocation circuit 5, the quantization bit length of the mantissa data is determined based on the exponent data input from the block floating point conversion circuit 4 by using auditory masking, and the determined quantization bit length is determined. The length is input to the unequal interval quantization encoding circuit 9.

In the non-equidistant quantization coding circuit 9, the non-equidistant quantization coding table shown in the following Table 1 (actually, it is stored as table information in the non-equidistant quantization coding circuit 9). Is used to perform the quantization coding process of the mantissa data according to the information of the quantization bit length of the mantissa data input from the bit allocation circuit 5 and the information of the band to which the mantissa data belongs. . Note that Table 1 will be described in detail later.

[0044]

[Table 1]

However, since a code in which all bits are "1", that is, a code of "11" or "111" is used for other identification, these are used as prohibited codes in this conversion. In addition, in cases other than Table 1 (the quantization bit length of the mantissa data is 2 or 3, and the coefficient data in one band is 1, 2 or 4)
(In the case of other than the above), equal-interval quantization coding is performed as in the conventional case.

When the quantized coding of the mantissa data is performed by the unequal interval quantization coding circuit 9 as described above,
The two data output terminals 7 and 8 output exponent part data and quantized coded data of mantissa part data, which are accumulated in a recording medium (not shown) or transmitted via a transmission path.

Next, the mantissa data output from the block floating point conversion circuit 4 is quantized and encoded by the unequal interval quantization encoding circuit 9 using the unequal interval quantization encoding rules shown in Table 1. I will explain how to do.

For example, when the quantization bit length of the mantissa data input from the bit allocation circuit 5 is 2 and the coefficient data in one band is 1, the mantissa input from the block floating point conversion circuit 4 Part data is from +0.5 to +
If it is between 1, the unequal interval quantization coding circuit 9 is
Then, the code "01" is output. When this code “01” is decoded, it represents 0.75 as mantissa data.

Similarly, if the mantissa data input from the block floating point conversion circuit 4 is 0, the code is "0".
At 0 ", the mantissa data after decoding becomes 0. Also, if the input mantissa data is between -1 and -0.5, the code is" 10 "and the mantissa after decoding. Part data is −
It becomes 0.75.

Further, when the quantization bit length of the mantissa data input from the bit allocation circuit 5 is 2 and the coefficient data in one band is two, the "mantissa part before quantization encoding" in Table 1 is performed. If the mantissa data input from the block floating point conversion circuit 4 is between +0.5556 and +1 by referring to the column of "Data", the unequal-interval quantization coding circuit 9 follows the code "01" according to Table 1. Is output.

Further, when the quantization bit length of the mantissa data input from the bit allocation circuit 5 is 3 and the coefficient data in one band is two, the "mantissa part before quantization coding" in Table 1 is shown. If the mantissa data input from the block floating point conversion circuit 4 is between +0.80952 and +1 by referring to the column of “Data”, the unequal interval quantization encoding circuit 9
The code “011” is output according to Table 1.

Similarly, the mantissa data input from the block floating point conversion circuit 4 is from +0.61905 to +.
If it is between 0.80952, the unequal interval quantization encoding circuit 9 outputs the code “010” according to Table 1. Similarly, the corresponding code is output according to which section of the 7-divided section the mantissa data input from the block floating point conversion circuit 4 belongs.

In other cases as well, the unequal interval quantization coding circuit 9 inputs the quantization bit length information of the mantissa data input from the bit allocation circuit 5 and the block floating point conversion circuit 4 in the same manner. The table determined as shown in Table 1 is referred to based on the information on the band to which the mantissa data belongs. Then, a code is selected and output according to which range the input mantissa data belongs.

Next, a method of creating the unequal interval quantization coding table shown in Table 1 will be described. For example, when the quantization bit length of the mantissa data is 2 and the coefficient data in one band is 1, the probability of occurrence of the mantissa data is +0.5 to +1.
And even between -1 and -0.5. Therefore, with respect to the mantissa data before decoding between +0.5 and +1, the average value 0.75 of the section is used as the mantissa data after decoding, and from -1 to -0.5.
With respect to the mantissa data before decoding during the period, the average value −0.75 of the section is set as the mantissa data after decoding.

The quantization bit length of the mantissa data is 2
When the number of coefficient data in one band is two, the probability of occurrence of the mantissa data is between -1 and -0.5, between -0.5 and +.
Between 0.5 and 3/8 between +0.5 and +1 respectively,
It becomes 1/4 and 3/8. The probability of occurrence of each of these sections is -1
3 divisions with the same probability of occurrence from 1 to +1 (when the quantization bit length is 2, use of the code “11” is prohibited, so there are 3 types of “00” “01” “10”). (Since the code is assigned, it is divided into three). When it is redone, the new section is from -1 to -0.55556, -0.5.
The interval from 556 to +0.5556 and +0.5556
It becomes a section from +1.

When the average value of each section is calculated from the occurrence probability of each section, -0.77778,0,
This is 0.77778, and these are the mantissa data after decoding. That is, when the code “10” is sent at the time of decoding, if the quantization bit length of the mantissa part data determined from the exponent part data is 2 and the coefficient data in one band is two. , The mantissa data is -0.77778
If it is decrypted as, it can be restored to the original data. Similarly, when another code "00" or "01" is sent, the original data can be restored by decoding the mantissa data as 0 or 0.77778.

The quantization bit length of the mantissa part data is 3
Then, when there are four coefficient data in one band, the probability of occurrence of the mantissa data is between -1 and -0.5, between -0.5 and +.
5/1 between 0.5 and +0.5 to +1
It becomes 6, 3/8, 5/16. It is divided into 7 so that the occurrence probability of each of these sections is the same between -1 and +1 (when the quantization bit length is 3, use of the code "111" is prohibited, so "111" is Since seven codes other than the above are assigned, the number of divisions is seven.) When re-doing, the new section is a section from -1 to -0.77143, -0.77.
143 to -0.54286 section, -0.54286
From -0.19048, from -0.19048 to +
0.19048 section, +0.19048 to +0.5
4286 section, +0.54286 to +0.7714
It becomes the section of 3 and the section of +0.77143 to +1.

Then, when the average value of each section is calculated from the occurrence probability of each section, -0.88571, -0.
65714, -0.38095, 0, 0.38095,
0.65714 and 0.88571, and the average values of these are coded as "100", "101", "110", "0", respectively.
00 "" 001 "" 010 "" 011 "is the mantissa data when decrypted.

In cases other than the above, for example, when the quantization bit length of the mantissa data is 2, assuming that the number of coefficient data in one band is n, then between -1 and -0.5. ,
The occurrence probabilities of the mantissa data between −0.5 and +0.5 and between +0.5 and +1 are (n + 1) / 4n,
(N-1) / 4n and (n + 1) / 4n. Therefore,
Re-division is performed so that the occurrence probability of each of these sections is the same between -1 and +1. Then, an average value is calculated from the occurrence probabilities of the respective sections, and these are used as the decoded mantissa data. Quantization bit length of mantissa data is 4
Also in the above cases, the conversion table can be created according to the same rule.

In this embodiment, the quantization bit length of the mantissa data as shown in Table 1 is 2 or 3,
When the number of coefficient data in one band is other than 1, 2 or 4, equidistant quantization coding is performed. Of course, it is also possible to perform the above-mentioned unequal interval quantization coding in all cases.

Next, decoding of audio data will be described. In the decoding device shown in FIG. 1B, the exponent part data and the quantized coded data of the mantissa part data recorded on a recording medium (not shown) or transmitted on the transmission path are input to the data input terminals 10 and 11, respectively. Is input to the bit allocation circuit 12 and the unequal-interval inverse quantization / decoding circuit 18 via.

In the bit allocating circuit 12, the quantizing bit length of the mantissa part data is determined based on the exponent part data input from the data input terminal 10 by using auditory masking, as in the case of encoding. Then, the determined quantization bit length is input to the unequal interval inverse quantization decoding circuit 18.

The unequal interval dequantization decoding circuit 18 quantizes and encodes the mantissa data input from the data input terminal 11 and the quantized bit length of the mantissa data input from the bit allocation circuit 12. And the band to which the mantissa data belongs, the mantissa data after decoding is extracted from the unequal interval quantization encoding table shown in Table 1, and is output to the block floating point inverse conversion circuit 14.

For example, when the quantization bit length of the mantissa part data is 3 and the coefficient data in one band is 4, and "110" is input as the quantized coded data of the mantissa part. The unequal interval inverse quantization decoding circuit 18 outputs -0.38905 as the decoded mantissa data to the block floating point inverse conversion circuit 14.

In the block floating point inverse conversion circuit 14,
The exponential part data input from the data input terminal 10 and the mantissa part data input from the unequal interval inverse quantization decoding circuit 18 are converted into floating point or fixed point coefficient data, and this is converted to IMDCT (Modified Discrete Cosine Inverse). It is input to the conversion circuit 15.

In the IMDCT circuit 15, the coefficient data input from the block floating point inverse conversion circuit 14 is subjected to inverse MDCT processing in block units, for example, 512 samples as one block, and the processing result is subjected to the windowing circuit 16 Entered in. In the windowing circuit 16, windowing processing is performed for each block, and 50% of the data of the current block is overlap-added to the data of the previous block. The audio data thus obtained is output from the audio data output terminal 17.

In the present embodiment, the code in which all the bits are "1", that is, the codes "11" and "111" are prohibited codes in the floating point conversion for use in other identification. If is not a prohibition code, the same number of codes can be assigned to the mantissa part on the positive side and the negative side centered on zero. Therefore, it is not necessary to assign a specific code (code in which all bits are "0") to the occurrence probability distribution of zero.

[0068]

As described above, according to the present invention, since the quantization coding and the inverse quantization decoding of the audio data are performed according to the actual occurrence probability distribution of the mantissa data, the occurrence probability is small. The inconvenience of assigning a large number of codes to data can be eliminated, and the quantization error can be reduced. For this reason, noise due to quantization can be reduced, and audible sound quality can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of an audio data encoding device and a decoding device using MDCT which is an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of band division of MDCT coefficients.

FIG. 3 is a block diagram showing a configuration example of a conventional audio data encoding device and decoding device using MDCT.

[Explanation of symbols]

 1 audio data input terminal 2 windowing circuit 3 MDCT circuit 4 block floating point conversion circuit 5 bit allocation circuit 6 equidistant quantization coding circuit 7 data output terminal 8 data output terminal 9 non-equidistant quantization coding circuit 10 data input Terminal 11 Data input terminal 12 Bit allocation circuit 13 Equal interval inverse quantization decoding circuit 14 Block floating point inverse conversion circuit 15 IMDCT circuit 16 Windowing circuit 17 Audio data output terminal 18 Unequal interval inverse quantization decoding circuit

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] June 20, 1995

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] 0025

[Correction method] Change

[Correction content]

[0025]

Since the audio data encoding device of the present invention comprises the above technical means, the audio data is quantized and encoded according to the probability distribution of the values actually generated in the range of the mantissa data. Like For example, according to the invention described in claim 2, based on the information of the quantization bit length of the mantissa part data and the information of the frequency band to which the mantissa part data belongs, according to the actual occurrence probability distribution of the mantissa part data. The determined table is referred to, and the mantissa data is quantized and coded according to this table. As a result, noise due to quantization can be reduced, and audible sound quality is improved.

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0026

[Correction method] Change

[Correction content]

Further, according to the audio data decoding apparatus of the present invention, the inverse quantization decoding of the audio data is performed according to the probability distribution of the values actually generated in the range of the mantissa data. become. For example, according to the invention described in claim 6, based on the information of the quantization bit length of the mantissa data and the information of the frequency band to which the mantissa data belongs, according to the actual occurrence probability distribution of the mantissa data. The determined table is referred to, and according to this table, the quantized and encoded mantissa data is dequantized and decoded, and the error due to quantization is reduced.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0028

[Correction method] Change

[Correction content]

The probability distribution that the mantissa data is generated is
Since it changes according to the number of coefficient data in one frequency band, noise due to quantization can be reduced, and the sound quality in hearing can be improved.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0031

[Correction method] Change

[Correction content]

However, in practice, -0.5 to +0.5
The probability of occurrence of mantissa data in the range is 0, and the average value on the negative side is -0.75, which is the average value in the range from -1 to -0.5. Therefore, if -0.667 is used as the mantissa data and is multiplied by the exponent part data to decode, then -0.75 is used as the mantissa part data and is decrypted by multiplying the exponent part data. Since it is biased, noise due to quantization is large.

[Procedure Amendment 5]

[Document name to be amended] Statement

[Name of item to be corrected] 0032

[Correction method] Change

[Correction content]

On the other hand, in the present embodiment, in accordance with the actual occurrence probability of the mantissa data, the range of possible values of the mantissa data is subdivided into sections so that the occurrence probabilities are equal. Since the code is assigned and quantized and encoded, the noise due to the above-described quantization can be reduced.

Claims (8)

[Claims] 1. An audio data encoding device for orthogonally transforming audio data on a frequency axis and encoding coefficient data obtained by dividing the data into exponent part data and mantissa part data, the mantissa part data The range of values that can be taken by is divided into a plurality of sections at unequal intervals according to the probability distribution in which the mantissa data actually occurs, and codes are assigned to the sections obtained by the plurality of divisions. An audio data encoding device, characterized in that it has unequal interval quantization encoding means for performing encoding. 2. A windowing means for windowing input audio data, an orthogonal transforming means for orthogonally transforming the data windowed by the windowing means to obtain coefficient data, and the orthogonal transforming means. Block floating point conversion means for obtaining the exponent part data and mantissa part data for each frequency band by block floating point conversion of the coefficient data obtained by the above, and the exponent part data obtained by the block floating point conversion means. Based on the probability distribution in which the mantissa data is generated, the bit allocation means for allocating the quantized bit length of the mantissa data based on the table, and the mantissa data of the mantissa data allocated by the bit allocation means. The table is searched based on the quantization bit length information and the frequency band information to which the mantissa data belongs. Having a non-equidistant quantization coding means for performing quantization encoding of the mantissa data audio data encoding apparatus according to claim. 3. The table according to the quantization bit length of the mantissa part data is divided into a plurality of sections in which the value range of the mantissa part data is divided so that the occurrence probabilities of the mantissa part data are equal. The audio data encoding device according to claim 2, wherein the audio data encoding device is determined by allocating a plurality of codes determined by 4. The audio data encoding device according to claim 3, wherein in the table, a code of all bits of zero is assigned to a section where the probability of occurrence of the mantissa data is zero. 5. An audio data decoding device for decoding audio data, wherein coefficient data obtained by orthogonal transformation is divided into exponent part data and mantissa part data to decode encoded audio data, wherein the mantissa part data is actually An audio data decoding apparatus comprising unequal interval dequantization decoding means for performing dequantization decoding of audio data coded according to the probability distribution generated in 1. 6. A bit allocation means for allocating a quantized bit length of mantissa data based on input exponent data, and a table determined according to a probability distribution in which the mantissa data is generated. The table is searched based on the quantized bit length information of the mantissa data allocated by the allocating means and the information of the frequency band to which the mantissa data belongs, and the quantized and encoded mantissa data is reversed. Unequally spaced inverse quantized decoding means for quantized decoding, and block floating for converting mantissa part data and exponent part data inversely quantized and decoded by the unequal interval inverse quantized decoding means into coefficient data. The decimal point inverse transforming means, the inverse orthogonal transforming means for inverse orthogonally transforming the coefficient data transformed by the block floating point inverse transforming means, and the inverse orthogonal transforming circuit. Audio data decoding apparatus characterized by having a windowing unit for performing windowing in the inverse orthogonal transformed data. 7. The table according to a quantization bit length of the mantissa part data is divided into a plurality of sections in which a value range of the mantissa part data is divided so that the occurrence probabilities of the mantissa part data are equal. The audio data encoding device according to claim 6, wherein the audio data encoding device is determined by allocating a plurality of codes respectively determined by the above. 8. The audio data encoding device according to claim 7, wherein in the table, a code of all bits of zero is assigned to a section where the probability of occurrence of the mantissa data is zero.