JP2003316394A - System, method, and program for decoding sound - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a sound decoding system for generating a decoded sound with high sound quality, and also to provide its method, and its program. <P>SOLUTION: The system includes: a flatening means (104); a copying means (105); and an envelope adding means (106), and comprises a function to generate the frequency signal of a certain frequency band by flatening a spectrum approximate form concerning a part of a frequency area signal which is decoded by a conventional decoding means, copying it to another frequency band, and adding the proper spectrum approximate form to the copied flat signal. The code amount of information about which kind of spectrum approximate form is added is generally smaller than the code amount which is generated by Huffman encoding in the frequency band, so that efficiency is raised in encoding. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to a speech decoding system,
The present invention relates to a voice decoding method and a voice decoding program (hereinafter referred to as a voice decoding system), and more particularly to a voice decoding system capable of generating decoded voice with high sound quality.

[0002]

2. Description of the Related Art A conventional audio decoding system will be described by taking the MPEG-2 AAC system which is an international standard audio encoding system as an example. For details of the operation of MPEG-2 AAC, refer to the standard document "Information Technology-Generic Coding of Moving Pictures and Associated Audio, 1997, Part 7: Advanced.
Audio Coding, AC ”(Info
rmation Technology-Generi
c coding of moving picture
es and associated audio,
Part7: Advanced Audio Codi
ng, AAC ”and is widely known to those skilled in the art. Here, the outline of the operation of the decoding system will be described with reference to FIG.

A conventional audio decoding system represented by MPEG-2 AAC has a bit stream demultiplexing means 5.
00, Huffman decoding means 501, dequantization means 502,
It comprises an inverse mapping conversion means 503.

Bitstream demultiplexing means 500
Separates the information multiplexed in the input bit stream, and Huffman decoding means 501 and inverse quantization means 5
02, outputs information required for each processing to the inverse mapping conversion means 503. For example, to the Huffman decoding unit 501, information regarding the Huffman-encoded quantized value data, the Huffman encoding unit used, and the like is output. The dequantization means 502 outputs a parameter indicating the quantization accuracy and the like, and is used when dequantizing each quantized value. The inverse mapping conversion unit 503 outputs a conversion function used in the inverse mapping conversion, a parameter representing a window coefficient, and the like.

The Huffman decoding means 501 Huffman-decodes the quantized value data input from the bitstream demultiplexing means 500 to generate a plurality of quantized values. The plurality of quantized values are signals in the frequency domain and represent the signal strength at each frequency. The generated plurality of quantized values are output to the inverse quantizer 502.

The dequantizing means 502 dequantizes a plurality of quantized values input from the Huffman decoding means 501 based on a parameter representing the quantization accuracy input from the bitstream demultiplexing means 500, To generate an audio signal in the frequency domain.

The inverse mapping conversion means 503 is an inverse quantization means 5
The frequency domain audio signal input from 02 is subjected to inverse mapping transformation based on the information about the transform function and the window coefficient input from the bitstream demultiplexing means 500 to generate a decoded audio signal in the time domain. Output.

[0008]

However, the quantized value output by the Huffman decoding means 501 and the audio signal in the frequency domain output by the inverse quantization means 502 have a value of zero, especially in a high frequency band. There are many. This occurs because the audio coding system cannot code these signals in the high frequency band and multiplex them into the bit stream due to the bit rate limitation. When the quantized value is compressed by the Huffman coding, it is impossible to compress the code amount to a very small amount due to the limit of the compression rate, and many frequency bands that cannot be coded occur mainly in the high frequency band. The frequency domain signal in such a frequency band has a value of zero in the speech decoding system. In this way, when the value of zero becomes high in the high frequency band,
The frequency band of the decoded voice signal obtained by the voice decoding system is narrowed, and the audible sound quality is deteriorated.

Therefore, the problem of the prior art is that the quality of the decoded speech signal is poor. The reason is that the frequency domain signal often has a value of zero in the high frequency band.

An object of the present invention is to provide a voice decoding system and a voice decoding method capable of generating a decoded voice signal with high sound quality.

[0011]

The speech decoding system of the present invention comprises at least a flattening means, a copying means and an envelope adding means, and operates so as to efficiently generate a wide band frequency domain signal.

That is, the Huffman decoding means generates a quantized value in the frequency domain of the audio signal from the input bit stream, and the dequantization means dequantizes the quantized value to obtain the audio signal in the frequency domain. Compared with a conventional speech decoding system for generating and converting the frequency domain speech signal into a time domain decoded speech signal by the inverse mapping transforming means, all or all of the frequency bands not decoded by the conventional speech decoding system or Flattening means for flattening the amplitude of the audio signal in the frequency domain to generate a flat signal in order to generate a partial frequency band signal in the extension band; and copying a part of the flat signal into the extension band And a envelope adding unit that adds a spectrum envelope to the frequency domain signal copied to the extension band to generate a frequency domain signal in the extension band. .

From the input bit stream,
The Huffman decoding means generates a quantized value in the frequency domain of the speech signal, the dequantization means dequantizes the quantized value to generate a speech signal in the frequency domain, and the inverse mapping transformation means generates the quantized value in the frequency domain. Compared with a conventional voice decoding system that converts a voice signal into a decoded voice signal in a time domain, a frequency domain signal in an extension band that is all or a part of a frequency band not decoded by the conventional voice decoding system is generated. In order to do so, a flattening unit that flattens the amplitude of the audio signal in the frequency domain to generate a flat signal, a copying unit that copies a part of the flat signal to the extension band, and a copying unit that copies the flat signal to the extension band. Noise adding means for adding noise to the frequency domain signal, and envelope adding means for adding a spectrum envelope to the frequency domain signal with the added noise to generate a frequency domain signal in the extension band. Ete will be configured.

[0014]

First, the concept of the present invention will be described. INDUSTRIAL APPLICABILITY The present invention is a method used in addition to a conventional speech decoding system to efficiently generate a frequency signal in a certain frequency band, and a part of the frequency domain signal decoded by the conventional decoding means, A frequency domain signal is generated by flattening the spectrum outline, copying it to another frequency band, and adding an appropriate spectrum outline to the copied flat signal. Information about which frequency band is copied to where, and what spectrum outline is added is included in the bitstream in advance. Generally, the code amount of these pieces of information is smaller than the code amount generated by the Huffman coding in this frequency band, so that the coding efficiency can be improved.

(First Embodiment) Next, a first embodiment of the present invention will be described.
Embodiments will be described in detail with reference to the drawings.

Referring to FIG. 1, according to the embodiment of the present invention, a bit stream demultiplexing means 100, a Huffman decoding means 101, an inverse quantization means 102, an inverse mapping conversion means 103, a flattening means 104, and a copying means 105. , Envelope adding means 106.

Compared with FIG. 5, the flattening means 1 is shown in FIG.
04, a copying unit 105, and an envelope adding unit 106 are added. Also, in order to output information necessary for these added processes, the bitstream demultiplexing means 10
0 is different in function from the bitstream demultiplexing means 500 of FIG. Regarding other processes, the Huffman decoding means 101, the dequantization means 102, and the inverse mapping conversion means 103 in FIG. 1 are the Huffman decoding means 50 in FIG.
1, inverse quantization means 502, and inverse mapping conversion means 503, respectively. Therefore, the changed bit stream demultiplexing unit 100, the added flattening unit 104,
The operation of the copying means 105 and the envelope adding means 106 will be described in detail. Note that here, MPEG2-AAC
The Huffman decoding means 101 is used as a means for generating a quantized value in order to explain the case of addition to the conventional speech decoding system using the method. However,
The Huffman decoding unit 101 may be another entropy coding / decoding unit such as an arithmetic coding / decoding unit, regardless of the essence of the present invention.

Bitstream demultiplexing means 100
Compared with the bitstream demultiplexing means 500 shown in FIG. 5, the information necessary for each processing is added to the flattening means 104, the copying means 105, and the envelope adding means 106 newly added in FIG. The difference is in the output. The contents of the output information will be described in the explanation of each process.

The flattening means 104 includes an inverse quantization means 102.
The audio signal in the frequency domain is input from. Further, the flattening information indicating the flattening method to be used may be input from the bitstream demultiplexing unit 100, which will be described later. The input audio signal in the frequency domain represents a so-called frequency spectrum, and its rough shape is called an envelope. The flattening means 104 flattens the input frequency-domain audio signal by dividing it by its envelope. The flat signal obtained by the flattening is output to the copying unit 105. Details of the flattening method will be described later.

For example, N output from the inverse quantizer 102
The sound signals in each frequency domain are MDCT (1 to N). In addition, MDCT (1 to
A signal obtained by flattening N) is defined as F (1 to N). This F (1 to N) is output to the copying means 105 as a flat signal.

A flat signal is input from the flattening unit 104, a frequency domain audio signal is input from the dequantization unit 102, and copy source frequency information and copy destination frequency information are input from the bitstream demultiplexing unit 100 to the copying unit 105. The copying unit 105 copies a predetermined number of signals of a part of the flat signal from the frequency represented by the copy source frequency information to the frequency represented by the copy destination frequency information of the audio signal in the frequency domain. The number to be copied is not a predetermined number, but the bitstream demultiplexing means 10
It may be a variable supplied from 0. The frequency-domain audio signal subjected to the copying process is output to the envelope adding unit 106.

That is, the copying means 105 has the above-mentioned MD.
CT (1 to N), F (1 to N), copy source frequency information, and copy destination frequency information are input. When the frequency information represented by the copy source frequency information is F1, the frequency information represented by the copy destination frequency information is F2, and the number of signals to be copied is F3, the copying means 105 MDCTs for M = 0 to (F3-1).
Substitute F (F1 + M) for (F2 + M). When a plurality of sets of copy source frequency information and copy destination frequency information are recorded in the bit stream, a plurality of F1, F2, F
Do this for each of the three sets. This means that signal copying from a flat signal is performed for a plurality of frequency bands. The MDCT (1-
NN) is output to the envelope adding means 106. Where N
N may be the same as or different from N described above. For example, N
When N is larger than N, a decoded voice signal having a wider band than that of the conventional decoding system can be obtained. In this way, when NN is different from N, the inverse mapping transform means 103 may use an inverse transform function or window coefficient whose transform block length is NN instead of N.

The envelope adding means 106 receives the frequency domain audio signal from the copying means 105 and the envelope information from the bit stream demultiplexing means 100. Envelope adding means 1
At 06, an envelope is added to the signal copied from the flat signal by the copying unit 105 based on the envelope information. The frequency domain audio signal thus obtained is inverse mapping conversion means 1
It is output to 03.

That is, with respect to the MDCT (1 to NN) input from the copying means 105, the envelope indicated by the envelope information is added to the frequency band indicated by the envelope information. For example, when the frequency information of the frequency band indicated by the envelope information is E1 to E2, the envelope coefficient is E (E1 to E2), and the envelope is added by multiplication with the envelope coefficient, the envelope adding means 1
The process performed in 06 is MDCT for M = E1 to E2.
(M) is multiplied by E (M). The MDCT (1 to NN) generated in this way is output to the inverse mapping conversion unit 103.

It is to be noted that, with respect to the envelope coefficient E (E1 to E2), if all of them are encoded in the bit stream as envelope information, the code amount becomes large. Therefore, as the envelope coefficient, only the values of the start point and the end point, that is, E (E1) and E (E2) are encoded in the bitstream, and the other envelope coefficients are E (E1) and E (E) in the decoding system. E
It is better to calculate by interpolation from 2). That is, E1
For E3 with <E3 <E2, E (E3) is E
(E1) × (E2-E3) ÷ (E2-E1) + E (E
2) × (E3−E1) ÷ (E2−E1) or the like.

Further, regarding the envelope coefficient E (E1 to E2), a value represented by a curve approximation coefficient such as a coefficient or a spline obtained by performing linear prediction analysis by regarding this as a time-series signal is acquired from the bit stream. May be generated. In addition to such linear prediction analysis and curve approximation such as a spline, difference information between these and the envelope coefficient E (E1 to E2) is acquired from the bit stream and added to this to obtain the envelope coefficient E (E1 to E1). E2) may be generated.

Next, the method of flattening by the flattening means 104 will be described in more detail. As described above, the flattening unit 104 flattens the input audio signal in the frequency domain by dividing it by the envelope. A method of flattening the audio signal MDCT (1 to N) in the frequency domain to obtain the flat signal F (1 to NN) will be specifically described. L = 1 to N
For N, the flat signal F (L) is obtained by the following procedure using the search range R.

(Procedure 1) MDCT (LR) to MDC
The representative value is calculated among the (2R + 1) signal values up to T (L + R). The representative value is, for example, the maximum amplitude value or the average amplitude value of the signal, and the representative value is the maximum absolute value MAX here. (Procedure 2) F (L) if the typical value MAX is zero
= 0 (Procedure 3) If the typical value MAX is not zero, F
(L) = MDT (L) ÷ MAX

Here, MAX, which is a representative value in the vicinity of the frequency (here, the maximum amplitude value), is regarded as the envelope of the spectrum, and the voice signal in the frequency domain is divided by the envelope. It should be noted that the search range R representing the range for searching the representative value may be a constant or a variable value. In the case of a variable value, it is necessary for the bitstream demultiplexing means 100 to acquire the information regarding the value from the bitstream and output it to the flattening means 104.

As a method of setting the search range R to a variable value,
The pitch of the voice signal in the frequency domain is calculated in the voice decoding system, or is calculated in the voice encoding system and encoded into a bit stream, and this is acquired by the method of decoding by the voice decoding system. There is a method of changing the search range R according to the pitch value.
As a method for calculating the pitch, a generally known method such as a method using the autocorrelation of the voice signal MDCT (1 to N) in the frequency domain can be used. The search range R is preferably equal to or wider than the pitch, and the search range R may be increased as the pitch increases and decreased as the pitch decreases.

Next, an example of copy source frequency information indicating which frequency band of the flat signal is copied and copy destination information indicating which frequency band of the audio signal in the frequency domain is copied will be described. First, the block length related to the copy source frequency information is defined as P1, and the block length related to the copy destination frequency information is defined as P2. Here, P1 and P2 may each be a predetermined constant, or may be the same as the above-mentioned pitch value. Further, the flat signal F (1 to N) is composed of P1 signals in order of frequency from low frequency, and the audio signal MDCT (1 to NN) in the frequency domain is composed of P2 signals in order of frequency from low frequency. Divided into blocks. That is, in the flat signal F (1 to N), the first block is F (1 to P1) and the second block is F (1 to P1).
(P1 + 1 to 2P1), ... And so on. Similarly, frequency domain audio signal MDC
In T (1 to NN), the first block is MDCT (1 to NN).
P2) The second block is MDCT (P2 + 1 to 2P
2), ..., It is divided into blocks.

In the copying means 105, the copy source frequency information B input from the bit stream multiplexing / separating means 100.
O, by using the copy destination frequency information BD, the signal from the BO-th block of the flat signal is converted to the BD of the frequency domain audio signal.
Copy to the second block. That is, the signal from the P1 × BOth signal from the low band of the flat signal is copied to the P2 × BDth signal from the low band of the audio signal in the frequency domain. At this time, considering the phase difference between the two, the copy source is set to P1 ×.
It may be P1 × BO + PH instead of the BOth. Here, PH represents a phase difference and may be supplied by the bitstream demultiplexing means 100,
It may be obtained as PH = (BD × P2) MOD P1. Here, A MOD B is a calculation for obtaining the remainder when A is divided by B.

Next, the overall operation of this embodiment will be described with reference to the flow charts of FIGS.

First, the dequantization means (102 in FIG. 1) outputs a frequency domain audio signal to be flattened to generate a flat signal (step 201 in FIG. 2, 104 in FIG. 1). Next, a part of the flattened signal is copied to generate an audio signal in the frequency domain (step 202, 105 in FIG. 1). Further, an envelope is added to the copied voice signal to generate a voice signal in the frequency domain (step 203, 106 in FIG. 1).
The frequency-domain audio signal generated in this way is input to the inverse mapping transformation (103 in FIG. 1).

Next, the effect of the first embodiment of the present invention will be described.

In this embodiment, the flattening means (1 in FIG. 1) is used.
04), a copying unit (105 in FIG. 1) and an envelope adding unit (106 in FIG. 1) are provided, and a frequency signal in a certain frequency band is
Part of the frequency domain signal decoded by the conventional decoding means,
It has a function of flattening with the spectrum envelope, copying it to another frequency band, and adding a proper spectrum envelope to the copied flat signal to generate it. The coding amount of information about which frequency band is copied to where, and what kind of spectrum envelope is added is generally smaller than the coding amount generated by Huffman coding in this frequency band, and thus the coding efficiency is improved. Can be improved.

(Second Embodiment) Next, the second embodiment of the present invention will be described.
Embodiments will be described in detail with reference to the drawings.

Referring to FIG. 3, according to the embodiment of the present invention, the bitstream demultiplexing means 300, the Huffman decoding means 301, the dequantization means 302, the inverse mapping conversion means 303, the flattening means 304, the copying means 305. , Envelope adding means 306 and noise adding means 307.

Compared with the first embodiment of the present invention shown in FIG. 1, the second embodiment of the present invention shown in FIG. 3 is different only in that a noise adding means 307 is added. Also, in order to output necessary information to the noise adding means 307, the bit stream demultiplexing means 300
However, there is a difference in function as compared with the bitstream demultiplexing means 100 of FIG. All other processing is equivalent to that of the first embodiment. Therefore, the operation of the changed bit stream demultiplexing means 300 and the added noise adding means 307 will be described in detail.

Bitstream demultiplexing means 300
Is compared with the bitstream demultiplexing means 100 of FIG. 1, and is added to the noise adding means 307 newly added in FIG.
The difference is that it outputs the information required for. That is, the noise information used by the noise adding unit 307 is acquired from the bit stream and output to the noise adding unit 307.

The noise adding means 307 includes a copying means 30.
5, an audio signal in the frequency domain is input, and noise information is input from the bitstream demultiplexing means 300. The noise information represents amplitude information such as the average amplitude or the maximum amplitude of the noise to be added and in which band the noise is added. The noise adding means 307 generates a random number based on this amplitude information, and generates this random number. Add to the frequency band specified by the noise information of the audio signal in the frequency domain. The noise-added frequency-domain audio signal is output to the envelope adding means 306.

Next, the overall operation of this embodiment will be described with reference to the flow charts of FIGS. 3 and 4.

First, the dequantizing means (302 in FIG. 3) outputs the audio signal in the frequency domain to be flattened to generate a flat signal (step 401 in FIG. 4, 404 in FIG. 3). Next, a part of the flattened signal is copied to generate an expanded speech signal in the frequency domain (step 402 in FIG. 4, 30 in FIG. 3).
5). Then, noise having a specified amplitude is added to a part of the frequency domain signal (step 403 in FIG. 4, 3 in FIG. 3).
07). Further, an envelope is added to a part of the frequency domain signal (step 404, 406 in FIG. 3). The frequency domain audio signal thus generated is subjected to the inverse mapping transformation (30 in FIG. 3).
It becomes the input of 3).

Next, the effect of the second embodiment of the present invention will be described.

In this embodiment, the flattening means (3 in FIG. 3) is used.
04), copying means (305 in FIG. 3), envelope adding means (306 in FIG. 3), and noise adding means (307 in FIG. 3), and a frequency signal of a certain frequency band is decoded by the conventional decoding means. A part of the frequency domain signal is flattened in its spectrum shape and then copied to another frequency band, noise is added to a part of the band, and an appropriate spectrum shape is added to the copied flat signal. It has a function to generate by doing. Since the code amount of information about which frequency band is copied to where, what kind of noise and spectrum outline is added is generally smaller than the code amount generated by Huffman coding in this frequency band, The efficiency of conversion can be improved.

Finally, as shown in the first and second embodiments of the present invention, the frequency signal of a certain band is efficiently generated by means such as copying based on the frequency signal of another band. A method for efficiently generating a stereo signal in a speech decoding system that operates will be described. As a conventional method for efficiently generating a stereo signal, the above-mentioned MPEG-2A is used.
Intensity stereo means described in AC standard documents are well known. In the intensity stereo means, in generating a stereo signal composed of two channels, left and right, a frequency domain signal common to the left and right is generated by a method similar to that of a monaural signal, and the spectrum of the left channel included in the bit stream is generated. It is obtained by decoding the envelope information and the spectrum envelope information of the right channel.
This spectrum envelope information represents a scaling factor, and a frequency domain signal for the left channel is generated by multiplying the frequency domain signal common to the left and right by the scaling factor represented by the spectrum envelope information for the left channel. Similarly, the frequency domain signal of the right channel is generated by multiplying the frequency domain signal common to the left and right by the multiplication factor represented by the spectrum envelope information of the right channel. When such an intensity stereo means is used in a voice decoding system for efficiently generating a frequency signal of a certain band based on a frequency signal of another band by means such as copying as in the present invention, It is also possible to generate a common frequency domain signal based on frequency signals of other bands by means of copying or the like, and then multiply each by the magnification represented by the spectrum envelope information of the left channel and the right channel to generate a stereo signal. Further, instead of generating a frequency domain signal common to the left and right, a left channel frequency domain signal and a left channel frequency domain signal are generated using the same left and right generation means, that is, the same copy source frequency information and copy destination frequency information for the left and right channels. May generate a frequency domain signal of the right channel from the flat signal of the right channel, and multiply these frequency domain signals by the scaling factors represented by the spectrum envelope information of the left channel and the right channel to generate a stereo signal.

The present invention also programs the configuration or procedure shown in the first and second embodiments,
It can also be implemented by a CPU provided in a computer or the like.

[0048]

The effect of the present invention is to generate a decoded voice signal with higher sound quality. The reason is that the audio signal in the frequency domain has a unit capable of decoding even if the required information amount is smaller than in the past and the coding efficiency is improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a flowchart showing an operation of the first exemplary embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

FIG. 4 is a flowchart showing the operation of the second exemplary embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a conventional embodiment.

[Explanation of symbols]

100, 300, 500 Bitstream demultiplexing means 101, 301, 501 Huffman decoding means 102, 302, 502 Dequantization means 103, 303, 503 Inverse mapping conversion means 104, 304 Flattening means 105, 305 Copying means 106 , 306 Envelope adding means 307 Noise adding means

   ─────────────────────────────────────────────────── ─── Continued front page    F-term (reference) 5D045 DA20                 5J064 AA01 AA02 BA09 BA16 BC16                       BC21 BD04

Claims (18)

[Claims] 1. A quantized value in the frequency domain of an audio signal is generated from an input bit stream by means such as Huffman decoding, and inverse quantization means inversely quantizes the quantized value to produce audio in the frequency domain. A signal is generated and compared with a conventional speech decoding system in which the inverse mapping transforming means transforms the frequency domain speech signal into a time domain decoded speech signal. Flattening means for flattening the amplitude of the audio signal in the frequency domain to generate a flat signal in order to generate a frequency domain signal in the extension band which is all or a part; and a part of the flat signal in the extension band Copy means for copying to the extension band, and envelope adding means for adding a spectrum envelope to the frequency domain signal copied to the extension band to generate a frequency domain signal in the extension band. Audio decoding system according to claim. 2. A quantized value in the frequency domain of a voice signal is generated from an input bit stream by means such as Huffman decoding, and the inverse quantization means inversely quantizes the quantized value to produce a voice in the frequency domain. A signal is generated and compared with a conventional speech decoding system in which the inverse mapping transforming means transforms the frequency domain speech signal into a time domain decoded speech signal. Flattening means for flattening the amplitude of the audio signal in the frequency domain to generate a flat signal in order to generate a frequency domain signal in the extension band which is all or a part; and a part of the flat signal in the extension band Copy means for copying to, a noise adding means for adding noise to the frequency domain signal copied to the extension band, and a spectrum envelope for adding the noise to the frequency domain signal Audio decoding system characterized in that it comprises an envelope adding means for generating a frequency domain signal of the extension band Te. 3. The voice decoding system according to claim 1, wherein the flattening means obtains the flat signal by dividing the voice signal in the frequency domain by its spectrum envelope. 4. The speech decoding according to claim 3, wherein, in the frequency domain speech signal, the spectrum envelope at each frequency is set to a representative value in the vicinity of each frequency in the frequency domain speech signal. system. 5. The speech decoding system according to claim 4, wherein the range in the vicinity of the frequency changes according to the pitch of the speech signal in the frequency domain. 6. The speech decoding system according to claim 4, wherein the representative value is either maximum amplitude or average amplitude. 7. An envelope added by the envelope adding means to a signal of each frequency in the audio signal in the frequency domain,
The calculation is performed by interpolation processing from two or more known envelopes in the vicinity of the frequency of each frequency.
7. The speech decoding system according to any one of 1 to 6. 8. The copying device according to claim 7, wherein the copying source frequency is derived from the copying source frequency information recorded in the input bit stream and the pitch of the audio signal in the frequency domain. A speech decoding system according to item 1. 9. A quantized value in the frequency domain of an audio signal is generated from an input bit stream, the quantized value is dequantized to generate a frequency domain audio signal, and the frequency domain audio signal is generated. In order to generate a frequency domain signal in an extension band, which is all or part of a frequency band not decoded by the conventional speech decoding method, compared with a conventional speech decoding method for converting a speech signal into a decoded speech signal in the time domain. In addition, the amplitude of the audio signal in the frequency domain is flattened to generate a flat signal, a part of the flat signal is copied to the extension band, and a spectrum envelope is added to the frequency domain signal copied to the extension band. And a frequency domain signal in the extension band is generated. 10. A quantized value in the frequency domain of a voice signal is generated from an input bit stream, and the quantized value is dequantized to generate a voice signal in the frequency domain. In order to generate a frequency domain signal in an extension band, which is all or part of a frequency band not decoded by the conventional speech decoding method, compared with a conventional speech decoding method for converting a speech signal into a decoded speech signal in the time domain. In, to flatten the amplitude of the audio signal in the frequency domain to generate a flat signal, copy a part of the flat signal to the extension band, add noise to the frequency domain signal copied to the extension band,
A speech decoding method, wherein a spectrum envelope is added to the noise-added frequency domain signal to generate the extension band frequency domain signal. 11. The speech decoding method according to claim 9, wherein the flat signal is obtained by dividing the speech signal in the frequency domain by its spectrum envelope. 12. The speech decoding according to claim 11, wherein, in the frequency domain speech signal, the spectrum envelope at each frequency is set to a representative value in the vicinity of each frequency in the frequency domain speech signal. Method. 13. The speech decoding method according to claim 11, wherein the range near the frequency changes according to the pitch of the speech signal in the frequency domain. 14. The representative value is either maximum amplitude or average amplitude, according to claim 12 or 13.
The voice decoding method described in. 15. An envelope to be added to a signal of each frequency in the audio signal in the frequency domain is calculated by interpolation processing from two or more known envelopes in the vicinity of the frequency of each frequency. The speech decoding method according to any one of claims 9 to 14. 16. The copy means of the audio decoding system according to claim 1, wherein the copy source frequency is derived from the copy source frequency information recorded in the input bit stream and the pitch of the audio signal in the frequency domain. Claim 9
15. The speech decoding method according to any one of 1 to 15. 17. A computer generates a quantized value in the frequency domain of an audio signal from an input bit stream, dequantizes the quantized value to generate an audio signal in the frequency domain, and outputs the quantized value in the frequency domain. Compared with a conventional speech decoding method for converting the speech signal of No. 1 into a decoded speech signal in the time domain, the frequency domain signal of the extension band which is all or a part of the frequency band not decoded by the conventional speech decoding method is To generate, the amplitude of the audio signal in the frequency domain is flattened to generate a flat signal, a part of the flat signal is copied to the extension band, and a spectrum envelope is added to the frequency domain signal copied to the extension band. Is added to generate a frequency domain signal in the extension band, and the processing is executed. 18. A computer generates a quantized value in the frequency domain of an audio signal from an input bit stream, dequantizes the quantized value to generate an audio signal in the frequency domain, and outputs the quantized value in the frequency domain. Compared with a conventional speech decoding method for converting the speech signal of No. 1 into a decoded speech signal in the time domain, the frequency domain signal of the extension band which is all or a part of the frequency band not decoded by the conventional speech decoding method is To generate, the amplitude of the audio signal in the frequency domain is flattened to generate a flat signal, a part of the flat signal is copied to the extension band, and noise is added to the frequency domain signal copied to the extension band. A speech decoding program, characterized in that the processing is executed by adding a spectrum envelope to the noise-added frequency domain signal to generate a frequency domain signal in the extension band. Beam.