JPH08307277A - Method and device for variable rate voice coding - Google Patents

Abstract

PURPOSE: To generate coding data in which best sound quality is reproduced at reproduction by selecting a method of optimum data distribution through the fluctuation of local coding rate based on an information quantity on an audible sense psychology in the voice coder. CONSTITUTION: An acoustic information quantity calculation device 19 calculates acoustic information of a frame acoustic characteristic signal 18 and decides a bit allocation signal 20 in the case of quantization according to the acoustic information quantity. A coding rate re-distributer 16 corrects a distribution of a coding rate in N-frames so that a mean value of the coding rate for the N frames and a designated rate signal 14 are equal to each other and provides an output of a new coding rate 15 for each frame. Bit allocation processing is conducted in the acoustic information quantity calculation device 19 according to the new coding rate to decide the allocation signal 20. The re-quantizer applies re-quantization to a digital audio signal 8 according to the allocation signal 20 and provides an output of a re-quantization signal 21.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to an audio encoding device for recording audio on a high density recording disk, and in particular encoded data capable of reproducing the best sound quality within a range of a fixed transmission rate and recording density on a recording medium. The present invention relates to a method and an apparatus for audio compression coding for generating a.

[0002]

2. Description of the Related Art An example of a method for performing high-efficiency coding with a constant sound quality is disclosed in Japanese Patent Laid-Open No. 4-192724.

[0003]

The above-mentioned conventional method is effective for speech voices due to the peculiarity of its processing method.
In a general audio signal, especially in a digital audio system aiming at a CD level, the effect is not exerted so much. Further, since the existing digital audio system using voice compression has a fixed transmission rate, the attack sound portion is significantly deteriorated due to the lack of the number of usable bits. Further, in a portion where the amount of voice information is small, an unnecessarily large data capacity is secured. Therefore, the data capacity is not effectively utilized, and there is room for improving the sound quality.

[0004]

According to the present invention, the amount of information in the psychoacoustic information of an input voice signal is calculated, and the local coding rate is determined according to the magnitude of the amount of information. Then, the local coding rate is controlled so as to be constant when the average is taken in a predetermined frame unit.

[0005]

The acoustic information amount calculating means estimates the acoustic information amount, which is the information amount in the psychoacoustic sense of the input digital audio signal. As the audio information amount increases or decreases for each coding unit, the coding rate distribution unit increases or decreases the coding rate for each coding unit. As a result, a high coding rate is assigned to a coding unit having a large amount of acoustic information, and a low coding rate is assigned to a coding unit having a small amount of acoustic information.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT FIG. 1 is an embodiment of the present invention and is a flow chart of an encoding process composed of seven processing steps. The processing will be described in order.

In processing step 1, a digital audio signal input process is performed for each frame which is a basic unit of the encoding process. Next, in a processing step 2, the digital audio signal is subjected to the first encoding process to calculate the acoustic information amount which is the psychoacoustic information amount contained in the digital audio signal in the frame. .

FIG. 2 shows an example in which the acoustic information amount is calculated by dividing into 32 frequency bands from 0 to 31.
FIG. 2A shows, for each band, the signal power level S (i) and the mask level M (i) which is a threshold level of a sound that cannot be heard due to an in-band mask or inter-band scattering. , S (i) and M (i), the magnitude of the difference SMR (i) is shown in the shaded area. Figure 2 (b) shows SMR
(I) is extracted for each band.
What is added up to max (sbmax does not have to be 31) is adopted as the acoustic information amount.

In processing step 3, the coded data obtained by the first coding process and the index number indicating the acoustic information amount are stored. In processing step 4, N (N is 2
It is determined whether or not data for the above (natural number) frames has accumulated. If not, the process proceeds to step 1 to perform input processing for the next frame. If accumulated, go to step 5. In processing step 5, the stored N frames of data are recalled. In step 6, it is determined whether the acoustic information amount has increased or decreased within N frames by the index number indicating the recalled acoustic information amount. Then, the coding rate for each frame is determined so that the coding rate also changes according to the acoustic information amount of each frame and the average coding rate between N frames becomes a predetermined designated rate. In the processing step 7, the second encoding process, which is the main encoding process, is performed in accordance with the encoding rate determined above to form a data stream.

FIG. 3 specifically shows an example in which the frame number corresponding to the time axis and the coding rate for each frame are adjusted as the amount of audio information increases or decreases. FIG. 3 (a)
Shows the change of the acoustic information amount for each frame, which is obtained according to the procedure shown in FIG. On the other hand, the coding rate also has the same rising and falling characteristics as shown in FIG. As a result, the sound quality will be constant, but usually due to the restrictions of the recording medium,
A constant recording rate is required. So, in this example, 1
The coding rate for each frame is adjusted so that 0 frame is defined as one audio group (N = 10), and the average coding rate within one audio group becomes the specified rate value.

By thus determining the coding rate in units of N frames in accordance with the change in the amount of audio information for each frame, the deterioration of the sound quality is averaged and the best data is obtained within the usable data capacity. It is possible to generate encoded data that can maintain the sound quality.

FIG. 4 is an example in which the amount of acoustic information is obtained by another calculation method, and the signal power S (i) of each band is set to the band sbm.
What is added up to ax (sbmax does not have to be 31) is adopted as the acoustic information amount. In this example, the sound pressure level for each band is simply estimated without considering the minimum audible limit and the band-to-band mask. reflect.

FIG. 5 shows an example in which the SMR (i) for each band obtained as shown in FIG. 2 is weighted differently for each band. Here, the real number K (i) such that K (i) is equal to or greater than K (i + 1) is used, and this K (i) and SM are used.
The product of R (i) is obtained for each band, and the product is calculated as band sbmax
The value obtained by adding up to (sbmax does not have to be 31) is adopted as the acoustic information amount.

If the band from 0 to 31 is obtained by equally dividing the highest frequency, one band has a larger bark width in the lower frequency range, which has a great influence on the psychology of hearing. In such a case, if the weighting of the example shown in FIG. 5 is performed, the sound information amount can be more accurately estimated.

FIG. 6 is a diagram in which the same weighting as in FIG. 5 is applied to the example of FIG. 4, and the product of K (i) and S (i) is obtained for each band, and this is calculated as the band sbmax ( sbmax
Is not required to be 31) and the sum is added as the acoustic information amount. As a result, the estimation on the psychoacoustic axis can be easily performed without performing the conversion on the bark axis.

FIG. 7 shows an embodiment of the encoding apparatus according to the present invention. Input the digital voice signal 8,
A coded bit stream 9 and a voice encoder 11 that outputs an index 10 indicating the amount of audio information, and a recording head 1.
2, a recording medium 13 such as an optical disk or a magnetic disk, a coding rate redistributer 16 which inputs the audio information amount index 10 and a designated rate signal 14 and outputs a modified coding rate index 15, and further a speech coder 11 Requantizer 1 for requantizing digital audio signal 8 as contents
7 and an audio information amount calculator 19 for calculating the psychoacoustic information amount of the digital audio signal 8, and an acoustic information amount calculator 19 for estimating a data capacity to be used, and a requantized signal 21 are input and an encoded bit stream 9 is output. It is composed of a frame configurator 22. The operation will be described in order.

First, the amount of acoustic information is estimated by the first encoding process. The digital audio signal 8 is input to the audio encoder 11 for each frame which is a basic unit of encoding. The acoustic information amount calculator 19 inputs the frame acoustic characteristic signal 18 which is the digital audio signal 8 of one frame or the spectrum obtained by time-frequency converting the digital audio signal 8 and outputs the acoustic information amount which is the information amount in psychoacoustic sense. Is calculated, and bit allocation for quantization is determined according to the amount of acoustic information so that the quantization noise level becomes equal to or lower than the undetectable level. Then, the bit allocation signal 20 and the audio information amount index 10 are output.

Then, a second encoding process (main encoding) of the same audio data is performed. The digital audio signal 8 is input to the audio encoder 11 for each frame which is a basic unit of encoding. The acoustic information amount calculator 19 calculates the acoustic information amount of the frame acoustic characteristic signal 18, and bit allocation 20 for quantization according to the acoustic information amount is performed.
To decide. Here, the coding rate redistributer distributes the coding rates within the N frames so that the average value of the coding rates determined in the first coding process among the N frames becomes equal to the designated rate signal 14. It is corrected and a new coding rate 15 for each frame is output. Bit allocation processing is performed in the acoustic information amount calculator 19 according to this new coding rate, and the allocation signal 20 is determined. Then, the requantizer requantizes the digital voice signal 8 according to the allocation signal 20 and outputs the requantized signal 21. The frame composer 22 composes the coded bitstream 9 from the requantized signal 21 and outputs it. The encoded bitstream 9 is recorded on the recording medium 13 by the recording head 12.

In this way, N is set in the first encoding process.
By grasping how the acoustic information amount in the frame changes, the encoding process is performed in the second encoding process at the encoding rate adapted to the acoustic information amount. As a result, the degree of deterioration of the sound quality is averaged, and at the same time, the bit stream 9 that can obtain the best sound quality within the limited data capacity can be generated.
Further, the recording medium 13 is not limited to the disk, and may be a tape or the like. Further, instead of recording on the recording medium, the data may be sent to the transmission path.

In comparison with the embodiment shown in FIG. 7, FIG. 8 is additionally provided with an N-frame voice memory circuit 23 for storing the input signal 8.

The digital audio signal 8 is input to the memory circuit 23, and an audio signal 25 for estimating the acoustic information amount and an audio signal 24 for main encoding are output. The acoustic information amount calculator 19 inputs the audio signal 25, calculates the acoustic information amount which is the information amount in psychoacoustic, and outputs the acoustic information amount index 10. The coding rate redistributer stores the audio information amount index 10 for N frames, and calculates the average value of the coding rates for N frames and the designated rate signal 14.
So that and become equal, and the acoustic information amount index 1
The coding rate 15 that increases or decreases according to the increase or decrease of 0 is determined.

After the coding rate of each of the N frames is determined, the audio signal 2 is subjected to the main encoding process.
4 is input to the requantizer 17, and the frame acoustic characteristic signal 18 is sent to the acoustic information amount calculator 19. The acoustic information amount calculator 19 calculates the acoustic information amount of the frame acoustic characteristic signal 18, and determines the bit allocation 20 for quantization in accordance with the acoustic information amount. At this time, the coding rate 15 is input in time with the audio signal 24, and the allocation 20 is determined under the condition of the coding rate 15. Then, the requantizer requantizes the audio signal 24 according to the allocation signal 20 and outputs the requantized signal 21. The frame composer 22 uses the requantized signal 21
To form and output the encoded bit stream 9. The encoded bitstream 9 is recorded on the recording medium disk 13 by the recording head 12.

In this way, the state of changes in the amount of acoustic information in N frames is grasped, and the encoding process is performed in the main encoding process at the encoding rate that changes according to the amount of acoustic information. As a result, the degree of deterioration of the sound quality is averaged, and at the same time, the bit stream 9 that can obtain the best sound quality within the limited data capacity can be generated.

As a result, since the estimation of the acoustic information amount and the main encoding can be alternately performed within a fixed time, it is possible to perform the encoding processing similar to that in FIG. 4 in a short time and with a smaller recording processing delay time.

FIG. 9 shows the coding rate redistributer 16 of FIG.
Instead of the margin amount MARG 28 and the acoustic information amount 26
And a coding rate determiner 27 for deciding a coding rate 29 for the frame so that a margin amount MARG 28 (MARG is a real number) can be additionally reserved for the originally required acoustic information amount 26. , An average value calculator 30 that outputs an average value AVE31 of the encoding rates 29 up to the present time, an average value AVE31, a predetermined upper limit average value HIGH32, and a lower limit average value LOW33 are input,
When AVE31 becomes HIGH32 or higher, MARG28
In this example, the margin adjustment circuit 34 is provided to control the value of the MARG 28 so that the value of the MARG 28 is increased when the AVE 31 becomes LOW 33 or less.

The digital audio signal 8 is input to the audio encoder 11 for each frame which is a basic unit of encoding. The acoustic information amount calculator 19 inputs the frame acoustic characteristic signal 18 which is the digital audio signal 8 of one frame or the spectrum obtained by time-frequency converting the digital audio signal 8 and outputs the acoustic information amount which is the information amount in psychoacoustic sense. 26 is calculated. The coding rate deciding unit 27 decides the coding rate 29 so that the margin amount 28 can be given to the acoustic information amount 26. According to the coding rate 29, the acoustic information amount calculator 19 determines the bit allocation 20 for quantization. Then, the bit allocation signal 20 and the audio information amount 26 are output. Then, the requantizer requantizes the digital voice signal 8 according to the allocation signal 20 and outputs the requantized signal 21. The frame composer 22 composes the coded bitstream 9 from the requantized signal 21 and outputs it. The encoded bitstream 9 is recorded on the recording medium disk 13 by the recording head 12.

In parallel with this, the average value calculator 30 calculates the average value AVE31 of the coding rate 29 up to the present time. In the margin adjuster 34, AVE31 is HIGH32
If it is above, reduce the value of MARG28,
Becomes LOW 33 or less, the value of the MARG 28 is increased and the MARG 28 is sent to the coding rate determiner 27.

In this way, the variable rate coding process can be performed almost in real time. Also, if the timing of the break of the voice group and the change of the acoustic information amount is bad, a voice group having an information amount of the upper limit or more and a voice group of the lower limit or less may be formed adjacently, but in this example, the average value Since it is calculated by adding up, it is possible to avoid such an inconvenience. Further, the MARG 28 does not have to be a positive number and may be a negative number. When MARG28 is negative, the absolute value of MARG28 indicates the degree of deterioration.
The larger G28 is, the higher coding rate is still selected.

FIG. 10 shows another example of the present invention, which comprises three steps of step 35, step 36, and step 37, and step 35 further includes step 38 and step 3.
It consists of 3 steps of 9 and step 40, and step 3
7 is a flowchart of the encoding process which is further composed of two steps 41 and 42. The processing will be described in order.

First, in process step 35, the first encoding process is performed. In process step 38, the first input process of the digital audio signal is performed for each frame which is a basic unit of the encoding process. Next, in processing step 39, the first encoding process is performed on the digital audio signal to calculate the acoustic information amount which is the psychoacoustic information amount included in the digital audio signal in the frame. .

In process step 40, the index indicating the amount of acoustic information by the first encoding process is stored. This step 38, step 39, processing step 4
0 is repeated for each frame.

After the first encoding process of the audio signal is completed, the encoding rate for each frame is determined according to the increase or decrease of the acoustic information amount for each frame by the acoustic information amount index stored in the processing step 36. To do.

Subsequently, in the processing step 37, the second encoding processing is performed. In the processing step 41, the second input processing of the digital audio signal is performed for each frame. In the processing step 42, the second encoding process, which is the main encoding process, is performed in accordance with the encoding rate for each frame determined above to form a data stream. The steps 41 and 42 are repeated for each frame.

By thus determining the coding rate in accordance with the change in the amount of acoustic information for each frame, it is possible to average the deterioration of sound quality and generate coded data that can maintain the best sound quality. Become.

[0035]

Since the coding rate can be determined according to the psychoacoustic information amount, the margin in the portion where the bit is not required is put in the portion where the bit is required. As a result, the distortion due to encoding can be averaged and minimized, and the best sound quality can be maintained.

[Brief description of drawings]

FIG. 1 is an example of a processing flow of a digital voice encoding method to which the present invention is applied.

FIG. 2 is an example of estimating SMR as the amount of acoustic information in the present invention.

FIG. 3 is an example of a change in coding rate for each frame when the present invention is applied.

FIG. 4 is an example of estimating signal power as the amount of acoustic information in the present invention.

FIG. 5 shows SMR for each band as the amount of acoustic information in the present invention.
This is an example of estimating the weighted product.

FIG. 6 shows an example of estimating the amount of acoustic information in the present invention by weighting signal power for each band.

FIG. 7 shows an embodiment of a digital speech coding apparatus to which the present invention is applied.

FIG. 8 is an example of a digital speech encoding apparatus to which the present invention is applied, in which an input data memory circuit is added.

FIG. 9 is an example of a digital voice encoding device to which the present invention is applied, which is an example of enabling real-time processing.

FIG. 10 is another embodiment of the processing flow of the digital voice encoding method to which the present invention is applied.

[Explanation of symbols]

 8 Digital Speech Signal 9 Encoded Bit Stream 10 Acoustic Information Amount Index 11 Speech Encoder 15 Modified Encoding Rate Index 16 Encoding Rate Redistributer 19 Acoustic Information Amount Calculator

Claims (12)

[Claims] 1. A digital audio signal is input by a predetermined coding unit, the acoustic information amount is calculated by estimating the psychoacoustic information amount of the digital audio signal, and the digital audio signal is calculated based on the acoustic information amount. An audio compression code for outputting an allocation signal indicating allocation of data capacity when compressing, regenerating a requantized signal from the digital audio signal according to the allocation signal, and forming an encoded bitstream from the requantized signal. In the encoding method, the encoding rate for the encoding unit is increased when the acoustic information amount for each encoding unit increases, and the encoding rate for the encoding unit is decreased when the acoustic information amount decreases. A speech coding method characterized by: 2. A digital audio signal is input by a predetermined coding unit, the acoustic information amount is calculated by estimating the psychoacoustic information amount of the digital audio signal, and the digital audio signal is calculated based on the acoustic information amount. An audio compression code for outputting an allocation signal indicating allocation of data capacity when compressing, regenerating a requantized signal from the digital audio signal according to the allocation signal, and forming an encoded bitstream from the requantized signal. In the encoding method, the coding rate for the coding unit is increased when the amount of acoustic information for each coding unit is increased, and the coding rate for the coding unit is decreased when the amount of acoustic information is decreased, and The average value of the coding rates is determined in advance in a speech group formed of P coding units (P is a natural number of 2 or more). Speech encoding method characterized by controlling the coding rate so that the value of the specified coding rate was. 3. The speech encoding method according to claim 1, wherein the amount of acoustic information is calculated by dividing an input digital speech into N bands (N is a natural number of 3 or more) on a frequency axis, and a band i is calculated. Signal power S within (i is from lowest range 0 to highest range N-1)
(I) and mask power M as a result of psychoacoustic analysis
The difference SMR (i) from (i) is added over a certain band j (j is 0 or more and N-2 or less) to a certain band k (k is 1 or more and N-1 or less). A voice encoding method, wherein the audio information amount is obtained as each of the encoding units. 4. The voice coding method according to claim 1, wherein the amount of acoustic information is calculated by dividing the input digital voice into N bands (N is a natural number of 3 or more) on the frequency axis, and a band i is calculated. Signal power S within (i is from lowest range 0 to highest range N-1)
(I) and mask power M as a result of psychoacoustic analysis
The difference between (i) and SMR (i) is obtained, and K (i) is K
A real number K (i) that is equal to or greater than (i + 1) is added to the SMR
K (i) × SMR (i), which is the product of (i), from a certain band j (j is 0 or more and N-2 or less), a certain band k (k is 1 or more and N-1 or less) A speech coding method, characterized in that a sum obtained up to the above is obtained as the acoustic information amount for each coding unit. 5. The speech coding method according to claim 1, wherein the amount of acoustic information is calculated by dividing an input digital speech into N bands (N is a natural number of 3 or more) on a frequency axis, and a band i is calculated. Signal power S within (i is from lowest range 0 to highest range N-1)
(I) a certain band j (j is 0 or more and N-2 or less)
To a certain band k (k is not less than 1 and not more than N-1) and obtained as the acoustic information amount for each of the encoding units. 6. The voice encoding method according to claim 1, wherein the amount of acoustic information is calculated by dividing an input digital voice into N bands (N is a natural number of 3 or more) on a frequency axis, and a band i is calculated. Signal power S within (i is from lowest range 0 to highest range N-1)
(I) is multiplied by a real number K (i) such that K (i) is K (i + 1) or more. K (i) × S (i) is a certain band j (j is 0 or more). A speech coding method, characterized in that a sum obtained from (N-2 or less) to a certain band k (k is 1 or more and N-1 or less) is obtained as the acoustic information amount for each coding unit. . 7. A digital audio signal is input by a predetermined coding unit, the acoustic information amount is calculated by estimating the psychoacoustic information amount of the digital audio signal, and the digital audio signal is calculated based on the acoustic information amount. An audio compression code for outputting an allocation signal indicating allocation of data capacity when compressing, regenerating a requantized signal from the digital audio signal according to the allocation signal, and forming an encoded bitstream from the requantized signal. In the digitization method, a certain amount of margin MARG with respect to the acoustic information amount.
(MARG is a real number) The coding rate for the coding unit is determined so as to secure an extra data capacity, the average value AVE of the coding rates up to the present time for the coding unit is calculated, and the average value is calculated. When AVE becomes lower than a predetermined low rate limit value LOW, the above-mentioned constant margin amount MARG is increased to obtain the above-mentioned average value AVE.
Is higher than a predetermined high rate limit value HIGH, the above-mentioned constant margin amount MARG is reduced. 8. An audio information amount calculating means for inputting a digital audio signal by a predetermined coding unit, estimating the psychoacoustic information amount of the digital audio signal, and outputting it as an audio information amount, and the audio information amount. Allocation means for outputting an allocation signal indicating allocation of data capacity when compressing the digital audio signal based on the above; requantization means for generating a requantized signal from the digital audio signal according to the allocation signal; In a voice compression encoding device including a stream forming means for forming an encoded bit stream from a quantized signal, a coding rate for the encoding unit is increased when the acoustic information amount for each encoding unit increases. , If it decreases, adjust the coding rate allocation so that the coding rate for the above coding unit is lowered. Speech coding apparatus comprising: a. 9. An audio information amount calculating means for inputting a digital audio signal by a predetermined coding unit, estimating the psychoacoustic information amount of the digital audio signal, and outputting the audio information amount as the audio information amount. Allocation means for outputting an allocation signal indicating allocation of data capacity when compressing the digital audio signal based on the above; requantization means for generating a requantized signal from the digital audio signal according to the allocation signal; In a voice compression encoding device including a stream forming means for forming an encoded bit stream from a quantized signal, a coding rate for the encoding unit is increased when the acoustic information amount for each encoding unit increases. , The coding rate for the above coding unit is lowered, and a predetermined P number ( Is a natural number greater than or equal to 2), and a coding rate allocating unit that adjusts the coding rate so that the average of the coding rates becomes a predetermined designated coding rate within a speech group configured of coding units. A speech coding apparatus comprising: 10. An audio information amount calculating means for inputting a digital audio signal by a predetermined coding unit, estimating the psychoacoustic information amount of the digital audio signal, and outputting the audio information amount as the audio information amount. Allocation means for outputting an allocation signal indicating allocation of data capacity when compressing the digital audio signal based on the above; requantization means for generating a requantized signal from the digital audio signal according to the allocation signal; In a voice compression encoding device including a stream forming means for forming an encoded bit stream from a quantized signal, a coding rate for the encoding unit is increased when the acoustic information amount for each encoding unit increases. , If it decreases, the coding rate for the above coding unit is lowered, and a predetermined P P is a natural number of 2 or more) A coding rate for setting a coding rate adjusted so that the average of the coding rates becomes a predetermined designated coding rate in a speech group formed of coding units. A speech coding apparatus comprising: a distribution unit and a coding rate conversion unit that reconfigures the coded bitstream into a coded bitstream according to an adjusted coding rate. 11. An audio information amount calculation means for inputting a digital audio signal by a predetermined coding unit, estimating the psychoacoustic information amount of the digital audio signal, and outputting it as an audio information amount, and the audio information amount. Allocation means for outputting an allocation signal indicating allocation of data capacity when compressing the digital audio signal based on the above; requantization means for generating a requantized signal from the digital audio signal according to the allocation signal; In a voice compression encoding device provided with a stream forming means for forming an encoded bit stream from a quantized signal, a predetermined P (P is a natural number of 2 or more) encoding units of the digital audio signal are accumulated. Data holding means,
When the audio information amount for each coding unit increases, the coding rate for the coding unit is increased, and when it decreases, the coding rate for the coding unit is decreased, and a predetermined P is set. A code that sets a coding rate adjusted so that the average of the coding rates becomes a predetermined designated coding rate within a speech group that is composed of (P is a natural number of 2 or more) coding units. And requantization means for extracting the digital audio signal from the data holding means in time with the adjusted coding rate output from the coding rate distribution means. A speech coder that operates. 12. An audio information amount calculating means for inputting a digital audio signal by a predetermined coding unit, estimating the psychoacoustic information amount of the digital audio signal, and outputting the audio information amount as the audio information amount. Allocation means for outputting an allocation signal indicating allocation of data capacity when compressing the digital audio signal based on the above; requantization means for generating a requantized signal from the digital audio signal according to the allocation signal; In a voice compression coding apparatus provided with a stream composing means for composing an encoded bit stream from a quantized signal, a certain margin amount MARG with respect to the acoustic information amount.
(MARG is a real number) Coding rate determining means for deciding the coding rate for the coding unit so as to secure an extra data capacity, and an average value of the coding rates up to the present time for the coding unit ( A coding rate average value calculating means for obtaining AVE) and the average value (AV
When E) becomes lower than a predetermined low rate limit value (LOW), the above-mentioned constant margin amount (MARG) is increased, and the average value (AVE) is set to a predetermined high rate limit value (HIGH). A voice coding apparatus, comprising: a margin adjusting means for reducing the above-mentioned constant margin amount (MARG) when the value becomes higher.