JPH09288498A - Voice coding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve quality of a reproduction by performing a bit assignment efficiently while utilizing a continuity time masking effect in a voice coding device reducing the redundancy included in voice and audio signals. SOLUTION: The compensation quantity of a masking threshold value is calculated by a simultaneous masking compensating means 13 and a continuity time masking compensating means 14 based on information of a spectrum envelope calculating means 11 and a power calculating means 12 and a masking threshold value 15a in which maskings of front and rear frames are concidered is finally determined in a masking threshold determining circuit 15 and after the assigning of a bit quantity is adaptively performed according to power of the value 16a obtained by dividing the spectra information 11a calculated in the spectrum envelope calculating meas 11 by the masking threshold value and then the quantization of a spectrum parameter is performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding apparatus for reducing redundancy included in speech and audio signals, and more particularly to a speech coding apparatus capable of improving coding efficiency by taking continuous masking into consideration.

[0002]

2. Description of the Related Art Japanese Unexamined Patent Publication No. 7-
Some are disclosed in Japanese Patent No. 261799. FIG.
The block diagram of the configuration of the speech coding apparatus is shown in FIG. Hereinafter, the configuration of a conventional speech encoding apparatus will be described with reference to the drawings. It is assumed that the input audio signal is first divided into a plurality of frequency bands by the band dividing means before being input to the block of FIG. The block of FIG. 5 encodes one of the frequency bands. In FIG. 5, reference numeral 51 is an analyzing means for calculating an orthogonal transform coefficient with a predetermined transform block length, and 52 is an FFT (Fas
Fourier transform means for calculating the power spectrum by (T Fourier Transform), 53 is a quantizing means for quantizing the orthogonal transform coefficient with a given bit amount, and 54 is bit allocation information generation for calculating the bit allocation amount for the orthogonal transform coefficient. Numeral 55 is a formatting means for multiplexing the quantized data and the bit allocation information and outputting it as encoded data.

In this device, an input signal is divided into bands and separated, and the process of FIG. 5 is performed in each band while utilizing the uneven distribution of power in each band. That is, the dynamic range is reduced by reducing the uneven distribution of energy in each band, and the amount of bits is assigned to each band according to the ratio of each power. Further, by performing orthogonal transformation on the data after the band division, energy concentration is further enhanced and quantization is performed to improve coding efficiency. References summarizing these techniques include the Acoustical Society of Japan, Vol. 51, No. 10, pp. 790-796, "Audio Coding" Akihiko Sugiyama.

The operation of this conventional speech coder will be described below. The input voice signal is converted into frequency domain parameters by the analyzing means 51 by using a discrete Cosine transform (hereinafter referred to as DCT) as an orthogonal transform. On the other hand, the bit allocation information generating means 54
Then, based on the power spectrum information of the audio data obtained by the Fourier transform means, find a masking threshold value that corrects the increase in the minimum audible value due to simultaneous masking, and only the frequency components exceeding this threshold value Allocate a bit amount. However, since the amount of information that can be assigned to one frame is constant (when the bit rate is constant), the total bit amount is fixed and the bit amount is proportionally distributed to the amount exceeding the masking threshold. The quantizing means 53 quantizes the corresponding orthogonal transform coefficient in accordance with the bit allocation amount for each frequency component thus obtained for each frame.

By performing bit allocation using such a masking threshold value, distortion is less perceptible perceptually, and coding efficiency is improved. About hearing masking, for example, "Introduction to Auditory Psychology" BCJ Moore
Written by Kengo Ogushi and detailed by Seishin Shobo.

[0006]

However, in the configuration such as the above-mentioned conventional speech coding apparatus, the bit considering the continuous masking (also referred to as non-simultaneous masking) which is the masking effect of the time-shifted signal components. No assignment can be made.

In view of the above problems, the present invention performs bit allocation in consideration of simultaneous masking and continuous masking, so that distortion due to coding is perceived in a voice or a rising portion of an audio signal whose energy changes abruptly. It is an object of the present invention to provide a speech coding apparatus which becomes difficult and which can improve the coding efficiency as a whole.

[0008]

In order to achieve this object, a speech coding apparatus according to a first aspect of the present invention comprises a spectrum information calculation means for obtaining spectrum information of an input signal in frame units, and a spectrum envelope of the input signal. Spectral envelope calculation means for obtaining the power of the input signal in frame units, power calculation means for obtaining the power of the input signal in frame units, and a correction amount of the minimum audible value of hearing considering the influence of simultaneous masking based on the spectrum envelope of the input signal And a simultaneous masking correction means for determining the correction amount of the minimum audible value of the hearing considering the influence of the continuous masking based on the temporal change of the power of the input signal over a plurality of frames and the envelope of the input signal. Based on the required continuous masking correction means, the simultaneous masking correction means, and the respective correction amounts obtained by the continuous masking correction means. Masking threshold value determining means for correcting the minimum audible value of hearing and determining a masking threshold value, and dividing the spectrum information by the masking threshold value and adaptively allocating bits according to the power of the spectrum information divided by the masking threshold value. And a spectrum information quantization means for quantizing the spectrum information divided by the masking threshold according to the bit allocation amount.

A speech coding apparatus according to a second aspect is the speech coding apparatus according to the first aspect, wherein the bit allocation control means is responsive to a change in the power of the spectrum information divided by the masking threshold over a plurality of frames. , Bits are adaptively assigned to each frame.

According to a third aspect of the present invention, there is provided a speech coding apparatus, comprising: an orthogonal transforming means for orthogonally transforming an input signal into frequency components in frame units to obtain orthogonal transform coefficients; , A simultaneous masking correction means for obtaining a correction amount of a minimum auditory audible value in consideration of the effect of simultaneous masking on the basis of a spectral envelope obtained from the orthogonal transform coefficient, and temporal power over a plurality of frames of the power of the input signal. A continuous masking correction means for obtaining a correction amount of the minimum audible value of the hearing considering the influence of the continuous masking based on the change and the spectral envelope obtained from the orthogonal transform coefficient, the simultaneous masking correction means, and the continuous masking. Masking threshold value determining means for correcting the minimum audible value of hearing based on each correction amount obtained by the correcting means, and determining a masking threshold value, The division means for dividing the orthogonal transformation coefficient by the masking threshold, the bit allocation control means for adaptively allocating bits according to the power of the orthogonal transformation coefficient divided by the masking threshold, and the division means for division. A spectrum parameter quantizing means for quantizing the orthogonal transform coefficient according to the bit allocation amount is provided.

According to another aspect of the speech coding apparatus of the present invention, a linear prediction analysis unit for linearly predicting an input signal on a frame-by-frame basis to obtain a linear prediction coefficient and a prediction residual, and the prediction residual is orthogonally transformed into frequency components. Orthogonal transformation means for obtaining the orthogonal transformation coefficient by means of the above, power calculation means for obtaining the power of the input signal on a frame-by-frame basis, and minimum auditory sense considering the influence of simultaneous masking based on the spectral envelope obtained from the linear prediction coefficient. Simultaneous masking correction means for obtaining the correction amount of the audible value, the influence of the successive masking is considered based on the temporal change of the power of the input signal over a plurality of frames and the spectral envelope obtained from the linear prediction coefficient. Continuous masking correction means for obtaining the correction amount of the minimum audible value of hearing, each correction amount obtained by the simultaneous masking correction means, and the continuous masking correction means Based on the masking threshold value determining means for correcting the minimum audible value of the auditory sense to determine the masking threshold value, the spectrum envelope obtained from the linear prediction coefficient is divided by the masking threshold value, and the spectrum envelope of the masking threshold value is divided. An inter-frame bit allocation control means for adaptively allocating bits to each frame according to a change in power over a plurality of frames; and a codebook having a representative vector code for vector quantizing the orthogonal transform coefficient. , A spectrum parameter quantizing means for determining an optimum representative vector code and gain information for the orthogonal transform coefficient using the codebook, and quantizing the gain information according to the bit allocation amount. It is equipped with it.

According to a fifth aspect of the present invention, there is provided a speech coding apparatus, wherein linear prediction analysis means for linearly predicting and analyzing an input signal in frame units to obtain a linear prediction coefficient and a prediction residual, and the prediction residual is orthogonally transformed into frequency components. Orthogonal transformation means for obtaining the orthogonal transformation coefficient by means of the above, power calculation means for obtaining the power of the input signal on a frame-by-frame basis, and minimum auditory sense considering the influence of simultaneous masking based on the spectral envelope obtained from the linear prediction coefficient. Simultaneous masking correction means for obtaining the correction amount of the audible value, the influence of the successive masking is considered based on the temporal change of the power of the input signal over a plurality of frames and the spectral envelope obtained from the linear prediction coefficient. Continuous masking correction means for obtaining the correction amount of the minimum audible value of hearing, each correction amount obtained by the simultaneous masking correction means, and the continuous masking correction means Based on the masking threshold value determining means for correcting the minimum audible value of the auditory sense to determine the masking threshold value, the spectrum envelope obtained from the linear prediction coefficient is divided by the masking threshold value, and the spectrum envelope of the masking threshold value is divided. In accordance with a change in power over a plurality of frames, an inter-frame bit allocation control unit that adaptively allocates bits to each frame, and a representative vector code for vector quantizing the orthogonal transform coefficient, respectively, A codebook size such as the number of the representative vectors and the number of dimensions, and a plurality of codebooks each having a different required bit amount corresponding to the codebook size, and a plurality of codebooks corresponding to the bit allocation amount from the plurality of codebooks. Select a codebook of codebook size and use the selected codebook. It has a configuration that includes a spectral parameter quantization means for determining the code and gain information of the optimum representative vector for the orthogonal transform coefficient Te.

A speech coding apparatus according to a sixth aspect is the speech coding apparatus according to the fourth aspect.
Alternatively, in the speech coding apparatus of No. 5, the spectrum parameter quantizing means is configured to determine the code and gain information of the representative vector on a perceptually weighted scale with respect to the frequency axis.

According to a seventh aspect of the present invention, there is provided a voice encoding device, which comprises a band dividing means for dividing a frequency band of an input signal into a first frequency band and a second frequency band lower than the first frequency band.
Band bit allocation control means for respectively allocating to the first and second frequency bands an amount of bits corresponding to the power ratio of signals in the respective frequency bands, and the signal in the first frequency band are input as input signals. The first speech coding module comprising the speech coding apparatus according to claim 4, wherein the bit allocation control means allocates the bits allocated to the first frequency band, and the second speech coding module. 6. The voice code according to claim 4, wherein a signal in the frequency band of 1 is input as an input signal, and the inter-frame bit allocation control means allocates the bits allocated to the second frequency band. And a second speech encoding module composed of an encoding device.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1. The first embodiment of the present invention is defined in claims 1, 2,
It corresponds to 3. FIG. 1 is a block diagram showing the configuration of the speech coding apparatus according to the first embodiment. In FIG. 1, 11 is orthogonal transformation means (spectrum information calculation means, spectrum envelope calculation means), 12 is power calculation means, 13 is simultaneous masking correction means, 14 is continuous masking correction means, 15 is masking threshold value determination means, 16 Is a division means, 17 is a spectrum parameter quantization means (spectrum information quantization means), and 18 is a bit allocation control means.

The operation will be described below. The orthogonal transform means 11 orthogonally transforms the input audio signal into frequency components in frame units by DCT or the like, and orthogonal transform coefficient 1
1a and the spectral envelope 11 of the orthogonal transform coefficient 11a
Find b. The power calculation means 12 calculates the power of each frame of the input audio signal. The simultaneous masking correction means 13 obtains the correction amount 13a of the minimum auditory audible value by simultaneous masking for each frame based on the spectral envelope 11b. Further, the continuous masking correction means 1
4 is a minimum audible value correction amount 14 by successive masking based on the power 12a for the past several frames obtained by the power calculation means 12 and the spectrum envelope 11b.
a is obtained for each frame. Here, it is also possible to input the orthogonal transform coefficient 11a to the simultaneous masking correction means 13 and the successive masking correction means 14 and obtain the spectrum envelope 11b by these correction means 13 and 14.

As the continuous masking, both anterograde masking in which the masker precedes the masky and a retrograde masking in which the masker appears after the masky can be considered. Now, if the frame length is set to about 10 msec, a masking effect is produced according to the power of the current frame and the power values of the frames before and after it. The amount of this mask differs depending on whether it is the forward masking or the backward masking. Further, the minimum audible value should be corrected by successive masking both before and after the frame to be corrected. Therefore, the power 12a for several frames and the value of the spectrum envelope 11b are temporarily stored in a memory or the like, and the correction amount 14a of the successive masking is determined based on the past frames. The masking threshold value determining means 15 determines the correction amount 13a provided by the simultaneous masking correcting means 13 and the successive masking correcting means 14 on the basis of the minimum audible value of hearing.
14a, the minimum audible value is corrected for each frame so as to be increased by the correction amounts 13a and 14a, and the corrected minimum audible value is output as the masking threshold 15a. Regarding simultaneous masking and continuous masking,
For example, see "Introduction to Auditory Psychology" above.

Next, in the dividing means 16, the orthogonal transformation coefficient 11a orthogonally transformed into the frequency component is divided by the masking threshold 15a of each frame finally decided by the masking threshold decision means 15 (subtraction on the logarithmic scale). Will be). The spectrum parameter quantizer 17 groups the divided orthogonal transform coefficients 16a for each certain bandwidth to form a predetermined frequency component group 17a, which is output to the bit allocation controller 18. The bit allocation control means 18 adaptively allocates bits to each frame according to the power value of each frequency component of the input predetermined frequency component group 17a, and at this time, the predetermined frequency component group Bits are adaptively assigned to each frame according to the change in the average power of each frame of 17a over a plurality of frames, and this is output to the spectrum parameter quantizing means 17 (18a).
At the same time, the auxiliary information 18b is output to the outside. Here, in the first embodiment, since the bit rate is fixed, the bit allocation for each frame is performed with the total bit amount of a plurality of frames being constant. The spectrum parameter quantizing means 17 quantizes the predetermined frequency component group 17a according to the bit amount 18a assigned to each frequency component of the predetermined frequency component group 17a of each frame, and encodes this. It is output to the outside as the output 17a.

As described above, in the first embodiment, the influence of the successive masking is also taken into consideration based on the temporal change of the power 12a of the input signal over a plurality of frames and the envelope 11b of the input signal. Since the masking threshold 15a is determined by the above, the distortion due to the coding is less likely to be perceived in the voice or the rising portion of the audio signal where the energy changes abruptly, and the coding efficiency can be improved as a whole.

Also, the bit allocation control means 18 divides the orthogonal transform coefficient 16a by the masking threshold 15a.
Since bit allocation is adaptively performed for each frame according to a change in average power of each frame of the (predetermined frequency component group 17a) over a plurality of frames,
When the orthogonal transform coefficient 16a after the division has an energy bias between a plurality of frames, the coding efficiency can be improved.

Further, the input signal is orthogonally transformed to obtain the orthogonal transformation coefficient 11a, the orthogonal transformation coefficient 11a is quantized, or the spectral envelope 11b is converted from the orthogonal transformation coefficient 11a.
Therefore, it is possible to provide a speech coder which can efficiently code an input signal by utilizing orthogonal transformation and also taking into account successive masking.

In the first embodiment, an example in which the orthogonal transform means is composed of DCT has been described. However, in addition to this, FFT,
It is also possible to configure by means of an orthogonal transformation system such as MDCT.

In the first embodiment, the coding is performed at a fixed bit rate, but the coding may be performed at a variable bit rate. In such a case, it is possible to perform bit allocation over the plurality of frames without limiting the total bit amount of the plurality of frames.

Embodiment 2 FIG. The second embodiment of the present invention corresponds to claims 1, 2, 4, and 6. FIG. 2 is a block diagram showing the configuration of the speech coding apparatus according to the second embodiment. In FIG. 2, 21 is a linear prediction analysis means (spectrum envelope calculation means), 22 is a power calculation means, and 23.
Is orthogonal transformation means (spectral information calculation means), 24 is simultaneous masking correction means, 25 is continuous masking correction means, 26 is masking threshold value determination means, 27 is a codebook having a code of a predetermined representative vector, 28
Is a spectrum parameter quantization means (spectrum information quantization means), and 29 is an inter-frame bit allocation control means.

The operation will be described below. The input speech signal is subjected to linear prediction analysis by the linear prediction analysis means 21 on a frame-by-frame basis to obtain a linear prediction coefficient and a prediction residual 21b, and further a spectrum envelope 21a is obtained from the linear prediction coefficient. The power calculation means 22 calculates the power of each frame of the input signal. The simultaneous masking correction means 24 is
Based on the spectral envelope 21a, a correction amount 24a of the minimum audible value of hearing due to simultaneous masking is obtained for each frame. Further, the successive masking correction means 25, based on the power 22a for the past several frames obtained by the power calculation means 22 and the spectral envelope 21a, corrects the minimum audible value correction amount 25a by the successive masking for each frame. Ask for. This continuous masking can be both anterograde masking in which the masker precedes the masky and a retrograde masking in which the masker appears after the masky. Now, if the frame length is set to about 10 msec, the successive masking of the current frame and the frames before and after the current frame will occur according to each power value.
The amount of this mask differs depending on whether it is the forward masking or the backward masking. Further, the minimum audible value should be corrected by successive masking both before and after the frame to be corrected. Therefore,
The power 22a for several frames and the values of the data of the spectrum envelope are temporarily stored in a memory or the like, and the correction amount 25a of the successive masking is determined in accordance with the past frames. The masking threshold value determining means 26 determines the correction amount 24 given by the simultaneous masking correcting means 24 and the successive masking correcting means 25 based on the minimum audible value of hearing.
In accordance with a and 25a, the minimum audible value is corrected for each frame so as to be increased by the correction amount 24a and 25a, and the corrected minimum audible value is output as the masking threshold value 26a. The orthogonal transform means 23 performs MDCT (Modified DC) on the prediction residual 21b obtained as a result of the linear prediction analysis.
The orthogonal transformation coefficient 23a is obtained by performing orthogonal transformation such as T).

Next, the inter-frame bit allocation control means 29 divides the spectral envelope 21a obtained by the linear prediction analysis means 21 by the masking threshold value 26a of each frame finally determined by the masking threshold value determining means 26. The spectrum envelope dF divided by the masking threshold value 26a is used when the bit allocation amount is distributed below, and the spectrum parameter quantizing means 28 is used.
It is also used as the weighting coefficient of the weighted distance measure when vector quantization is performed in. That is, the average power for a plurality of frames is calculated for the spectrum envelope dF after the division, and the bit allocation amount is distributed to each frame according to the ratio of the average power of each frame to the average power for the plurality of frames. And outputs the allocation of the bit allocation amount to the spectrum parameter quantizing means 28 together with the spectrum envelope dF after the division (29a). Then, the allocation of the bit allocation amount is supplemented by auxiliary information si (i)
And output to the outside. Here, in the second embodiment,
Since the bit rate is fixed, the bit allocation for each frame is performed with a constant total bit amount of a plurality of frames. Further, by considering the bit allocation over a plurality of frames, it is possible to improve the coding efficiency in the case where the spectrum envelope dF after division has a bias of energy among a plurality of frames.

Next, the spectrum parameter quantization means 2
8, the codebook 27 is used to generate the orthogonal transform coefficient 23 obtained by the orthogonal transform means 23 with the bit amount assigned to each frame by the bit assignment control means 29.
Vector quantization of a is performed. That is, the orthogonal transform coefficient 2
An optimum representative vector code 28a and gain information for 3a are determined, and the gain information is quantized according to the bit amount assigned to each frame (28b). Then, the code 28a and the quantized gain information 28b are output to the outside as a coded output. At this time, the code 28a of the representative vector is determined so as to minimize the distance measure weighted by the auditory sense using the weighting coefficient proportional to the spectral envelope dF after the division.

As described above, in the second embodiment, the linear prediction analysis of the input signal is performed to obtain the linear prediction coefficient and the prediction residual 21b, and the orthogonal transform coefficient 23a obtained from the prediction residual 21b is quantized. Since, or the spectrum envelope 21a is obtained from the linear prediction coefficient, it is possible to provide a speech coding apparatus that can efficiently code an input signal by using linear prediction analysis and also considering temporal masking.

Further, since the orthogonal transform coefficient 23a is vector-quantized by using the codebook 27 and the bits are adaptively assigned to the gain information when the vector quantization is performed, the coding efficiency can be further improved.

Further, the bit allocation for each frame is divided by the masking threshold to obtain a spectral envelope d.
Since it is adapted adaptively according to the change of the average power of F over a plurality of frames, it is possible to improve the coding efficiency in the case where the spectrum envelope dF after the division has an energy deviation among a plurality of frames. You can

Furthermore, when vector-quantizing the orthogonal transform coefficients, the code of the representative vector and the gain information are determined on a perceptually weighted scale with respect to the frequency axis. It is possible to configure an efficient voice encoding device.

In the second embodiment, an example in which the orthogonal transform means is composed of MDCT has been described.
It is also possible to configure by means of an orthogonal transformation system such as T or DCT.

In the second embodiment, the coding is performed at a fixed bit rate, but the coding may be performed at a variable bit rate. In such a case, it is possible to perform bit allocation over the plurality of frames without limiting the total bit amount of the plurality of frames.

Embodiment 3. The third embodiment of the present invention corresponds to claims 1, 2, 5, and 6. FIG. 3 is a block diagram showing the configuration of the speech coding apparatus according to the third embodiment. In FIG. 3, 31 is a linear prediction analysis means (spectrum envelope calculation means), 32 is a power calculation means, and 33.
Is orthogonal transformation means (spectral information calculation means), 34 is simultaneous masking correction means, 35 is continuous masking correction means, 36 is masking threshold value determination means, and 37 is a representative vector code. A codebook group consisting of a plurality of codebooks (0 to Nc) each having a different codebook size such as the number of dimensions and a required bit amount corresponding to the codebook size, and 38 is a spectrum parameter quantization means (spectrum information quantization means). ) And 39 are inter-frame bit allocation control means.

The operation will be described below. The input speech signal is subjected to linear prediction analysis by the linear prediction analysis means 31 on a frame-by-frame basis to obtain a linear prediction coefficient and a prediction residual 31b, and further a spectrum envelope 31a is obtained from the linear prediction coefficient. The power calculation means 32 obtains the power 32a of each frame of the input audio signal. The simultaneous masking correction means 34 uses the spectral envelope 31a as a basis.
Correction amount 34a of minimum audible value of hearing by simultaneous masking
Is calculated for each frame. Further, the successive masking correction means 35 includes the power 32a for the past several frames obtained by the power calculation means 32 and the spectrum envelope 31a.
Based on, the minimum audible value correction amount 3 by continuous masking
5a is obtained for each frame. This continuous masking can be both anterograde masking in which the masker precedes the masky and a retrograde masking in which the masker appears after the masky. Now, the frame length is 10 msec
To a degree, successive masking of the current frame with the previous and subsequent frames will occur depending on the value of each power. The amount of this mask differs depending on whether it is the forward masking or the backward masking. Further, the minimum audible value should be corrected by successive masking both before and after the frame to be corrected. Therefore, the power 32a for several frames and the values of the spectral envelope 31a are temporarily stored in a memory or the like, and the correction amount 35a of the successive masking is determined in the past frame. Masking threshold value determining means 36
On the basis of the minimum audible value of hearing, according to the correction amounts 34a, 35a provided by the simultaneous masking correction means 34 and the successive masking correction means 35, the minimum audible value of each frame is set to the correction amount 34a, 35a. It is corrected so as to be increased by the amount, and the corrected minimum audible value is output as the masking threshold 36a. The orthogonal transform means 33 uses the MDCT for the prediction residual 31b obtained as a result of the linear prediction analysis.
Orthogonal transformation such as is performed to obtain the orthogonal transformation coefficient 33a.

Next, the inter-frame bit allocation control means 39 divides the spectral envelope 31a obtained by the linear prediction analysis means 31 by the masking threshold value 36a of each frame finally determined by the masking threshold value determining means 36. The spectrum envelope dF divided by the masking threshold 36a is used in the following distribution of bit allocation amount, and the spectrum parameter quantizing means 38 is used.
It is also used as the weighting coefficient of the weighted distance measure when vector quantization is performed in. That is, the average power for a plurality of frames is calculated for the spectrum envelope dF after the division, and the bit allocation amount is distributed to each frame according to the ratio of the average power of each frame to the average power for the plurality of frames. Then, the distribution of the bit allocation amount is output to the spectrum parameter quantization means 38 together with the spectrum envelope dF after the division (39a). Then, the allocation of the bit allocation amount is supplemented by auxiliary information si (i)
And output to the outside. Here, in the third embodiment,
Since the bit rate is fixed, the bit allocation for each frame is performed with a constant total bit amount of a plurality of frames. Further, by considering the bit allocation over a plurality of frames, it is possible to improve the coding efficiency in the case where the spectrum envelope dF after division has a bias of energy among a plurality of frames.

Next, the spectrum parameter quantization means 3
8, the codebook group 37 is used to generate the orthogonal transform coefficient 3 obtained by the orthogonal transform means 33 with the bit amount assigned to each frame by the bit assignment control means 39.
Perform vector quantization of 3a. That is, from the plurality of codebooks (0 to Nc) of the codebook group 37,
A codebook having a required bit amount corresponding to the bit amount assigned to each frame, that is, a codebook size, is selected, and by using the selected codebook, a code 38a of an optimum representative vector for the orthogonal transform coefficient 33a is obtained. Gain information is determined, and the gain information is quantized with a fixed bit amount. Then, the code 38a and the quantized gain information 38b are output to the outside as a coded output. At this time, the representative vector code 38a
Is determined to minimize the perceptually weighted distance measure using a weighting factor proportional to the spectral envelope dF after division.

As described above, in the third embodiment, bits are adaptively assigned to each frame according to the change in the power of the spectrum envelope dF divided by the masking threshold value over a plurality of frames, Also, the orthogonal transform coefficient 33a is vector-quantized by selecting a codebook having a codebook size corresponding to the allocated bit amount from a plurality of codebooks (0 to Nc), and thus the spectrum after division is divided. When the envelope dF has a bias in energy among a plurality of frames, the coding efficiency can be further improved.

When the orthogonal transform coefficient 33a is vector-quantized, the code of the representative vector and the gain information are determined on the basis of the perceptually weighted scale with respect to the frequency axis. It is possible to configure an efficient voice encoding device.

In the third embodiment, an example in which the orthogonal transform means is composed of MDCT has been described.
It is also possible to configure by means of an orthogonal transformation system such as T or DCT.

Further, in the third embodiment, the coding is carried out at a fixed bit rate, but the coding may be carried out at a variable bit rate. In such a case, it is possible to perform bit allocation over the plurality of frames without limiting the total bit amount of the plurality of frames.

Embodiment 4 The fourth embodiment of the present invention corresponds to claim 7. FIG. 4 is a block diagram showing the configuration of the speech coding apparatus according to the fourth embodiment.
The fourth embodiment shows an example having a module that performs coding processing after dividing into two frequency bands. In FIG. 4, 41 is a band dividing means for dividing the frequency band of the input audio signal into a first frequency band and a second frequency band lower than the first frequency band, and 42 is power for calculating the power of the signal in each band. The calculating means 43 is a band bit allocation control means for allocating bits to each band. Reference numeral 44 is a first voice encoding module to which the signal of the first frequency band and the output of the band bit allocation control means 43 are input,
The speech coding apparatus according to the first embodiment is used. Reference numeral 45 is a second voice encoding module to which the signal of the second frequency band and the output of the band bit allocation control means 43 are input, and the voice encoding device of the second or third embodiment is used.

The operation will be described below. The band dividing means 41 divides the input audio signal into signals 41a and 41b in the first and second frequency bands and outputs the signals. The power calculation means 42 is the power 42a of the signal of each band after band division.
Then, the band bit allocation control means 43 determines the bit amounts 43a and 43b for each band in accordance with the ratio of the power 42a of the signal in each band. Here, by reducing the uneven distribution of energy in each band, it is possible to reduce the dynamic range and further improve the coding efficiency. The first and second speech coding modules 44 and 45 are
The signals 41a and 41b of the first and second frequency bands are encoded by the bit amounts 43a and 43b allocated by the band bit allocation control means 43, respectively.

Thus, the second method utilizing the linear prediction analysis
The low-frequency components that can be relatively linearly predicted are input to the speech coding module of 1), while the first speech coding module that uses orthogonal transform is likely to include a signal that is difficult to linearly predict. Since a large number of high frequency components are input, it is possible to increase the merit of using the first speech encoding module that uses orthogonal transformation. In the above description, the case where the frequency band is divided into two frequency bands has been described, but the first and second frequency bands may be further divided into a plurality of frequency bands.

In this way, the frequency band of the input voice signal is divided, and the voice coding module suitable for each frequency is used for each of the divided frequency bands to further improve the coding efficiency. realizable.

[0046]

As described above, according to the invention of claim 1, the spectrum information of the input signal is divided by the masking threshold value determined in consideration of the simultaneous masking based on the minimum audible value of hearing, and the division is performed. In a speech coding apparatus that quantizes subsequent spectrum information by adaptively allocating bits according to the power of the spectrum information, a temporal change in power of an input signal over a plurality of frames and an envelope of the input signal. Since the masking threshold value is determined in consideration of the effect of continuous masking based on the above, distortion due to coding is less likely to be perceived in a rising portion of a voice or an audio signal whose energy changes abruptly, A remarkable effect that the coding efficiency can be improved as a whole is obtained.

Further, according to the invention of claim 2, the bit allocation control means adaptively allocates a bit to each frame according to a change in the power of the spectrum information divided by the masking threshold value over a plurality of frames. Since this is done, it is possible to improve the coding efficiency when the spectrum information after the division has a bias in energy among a plurality of frames.

According to the invention of claim 3, the input signal is orthogonally transformed to obtain the orthogonal transform coefficient, and the orthogonal transform coefficient is quantized, or the spectrum envelope is obtained from the orthogonal transform coefficient. It is possible to provide a speech coding apparatus that can efficiently code an input signal by using orthogonal transformation and also considering successive masking.

Further, according to the invention of claim 4, the input signal is subjected to linear prediction analysis to obtain a linear prediction coefficient and a prediction residual, and the orthogonal transform coefficient obtained from the prediction residual is quantized, or the linear transformation coefficient is calculated. Since the spectrum envelope is obtained from the prediction coefficient, it is possible to provide a speech coder that can efficiently code the input signal by using the linear prediction analysis and also considering the temporal masking. In addition, since the orthogonal transform coefficient is vector-quantized using a codebook and bits are adaptively assigned to the gain information when the vector quantization is performed, the coding efficiency can be further improved. Furthermore, since the bit allocation for each frame is adaptively performed according to the change in the power of the spectrum envelope divided by the masking threshold value over a plurality of frames, the spectrum envelope after the division can reduce the energy of the plurality of frames. In the case where there is a bias, the coding efficiency can be improved.

Further, according to the invention of claim 5, bits are adaptively assigned to each frame according to a change in the power of the spectrum envelope divided by the masking threshold value over a plurality of frames, and a plurality of bits are assigned to each frame. Since a codebook having a codebook size corresponding to the allocated bit amount is selected from the codebooks and the orthogonal transform coefficient is vector-quantized, the spectrum envelope after division is biased in energy between a plurality of frames. It is possible to further improve the coding efficiency in the case of having.

Further, according to the invention of claim 6, the code of the representative vector and the gain information are determined by a perceptually weighted scale with respect to the frequency axis when vector-quantizing the orthogonal transform coefficient. It is possible to configure a speech coding apparatus with good coding efficiency that effectively uses masking.

Further, according to the invention of claim 7, the frequency band of the input signal is divided into a high frequency band and a low frequency band, and the signal of the high frequency band is subjected to orthogonal transformation. Input to a first speech coding module consisting of a coding device, and a signal in a low frequency band is inputted to a second speech coding module consisting of a speech coding device according to claim 4 or 5, which uses linear prediction analysis. As a result, a large number of relatively linearly predictable low frequency components are input to the second speech coding module that uses linear prediction analysis, while the first speech coding module that uses orthogonal transformation is used. A large number of high frequency components, which are likely to include signals for which linear prediction is difficult, will be input to, increasing the merit of using the first speech encoding module that utilizes orthogonal transform. It can, as a result, can achieve a further improvement in coding efficiency.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a speech coding apparatus according to Embodiment 1 of the present invention.

FIG. 2 is a block diagram showing the configuration of a speech coding apparatus according to Embodiment 2 of the present invention.

FIG. 3 is a block diagram showing the configuration of a speech encoding apparatus according to Embodiment 3 of the present invention.

FIG. 4 is a block diagram showing the configuration of a speech coding apparatus according to Embodiment 4 of the present invention.

FIG. 5 is a block diagram showing a configuration of a conventional speech encoding device.

[Explanation of symbols]

 11,23,33 Orthogonal transformation means 12,22,32 Power calculation means 13,24,34 Simultaneous masking correction means 14,25,35 Continuous masking correction means 15,26,36 Masking threshold value determining means 16 Division means 17,28 , 38 spectrum parameter quantization means 18 bit allocation control means 21 linear prediction analysis means 27 codebook 29, 39 interframe bit allocation control means 37 codebook group 41 band division means 43 band bit allocation control means 44 first speech coding Module 45 Second Speech Encoding Module

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Internal reference number FI Technical indication H03M 7/30 9382-5K H03M 7/30 A

Claims (7)

[Claims] 1. A spectrum information calculation means for obtaining the spectrum information of an input signal in frame units, a spectrum envelope calculation means for obtaining the spectrum envelope of the input signal in frame units, and a power calculation for obtaining the power of the input signal in frame units. Means, a simultaneous masking correction means for obtaining a correction amount of the minimum audible value of the hearing considering the influence of the simultaneous masking based on the spectral envelope of the input signal, and a temporal change of the power of the input signal over a plurality of frames. And a continuous masking correction means for obtaining a correction amount of the minimum auditory audible value in consideration of the influence of the continuous masking based on the envelope of the input signal, and the simultaneous masking correction means and the continuous masking correction means. The masking threshold that determines the masking threshold by correcting the minimum audible value of hearing based on each correction amount Determining means, the spectrum information is divided by the masking threshold, the bit allocation control means for adaptively bit allocation according to the power of the spectrum information divided by the masking threshold, the spectrum divided by the masking threshold A speech coding apparatus, comprising: spectral information quantizing means for quantizing information according to the bit allocation amount. 2. The speech coding apparatus according to claim 1, wherein the bit allocation control means adapts to each frame according to a change in power of the spectrum information divided by the masking threshold over a plurality of frames. A speech coding apparatus characterized in that bit allocation is carried out dynamically. 3. An orthogonal transformation means for orthogonally transforming an input signal into frequency components in frame units to obtain orthogonal transformation coefficients, a power calculation means for obtaining power of the input signal in frame units, and the orthogonal transformation coefficients. Simultaneous masking correction means for obtaining the correction amount of the minimum audible value of hearing considering the effect of simultaneous masking based on the spectrum envelope, and the temporal change of the power of the input signal over a plurality of frames and the orthogonal transform coefficient. A continuous masking correction means for obtaining a correction amount of the minimum audible value of the hearing in consideration of the effect of the continuous masking based on the spectral envelope, and the correction amounts obtained by the simultaneous masking correction means and the continuous masking correction means. Masking threshold value determining means for determining the masking threshold value by correcting the minimum audible value of the auditory sense based on Division means for dividing by a king threshold, bit allocation control means for adaptively allocating bits according to the power of the orthogonal transformation coefficient divided by the masking threshold, and the orthogonal transformation coefficient divided by the division means for the bit A speech coding apparatus, comprising: a spectrum parameter quantizing means for quantizing in accordance with an allocation amount. 4. A linear prediction analysis means for linearly predicting and analyzing an input signal on a frame-by-frame basis to obtain a linear prediction coefficient and a prediction residual, and an orthogonal transform for orthogonally transforming the prediction residual into a frequency component to obtain an orthogonal transform coefficient. A conversion means; a power calculation means for obtaining the power of the input signal in frame units; and a simultaneous calculation of a correction amount of the minimum auditory audible value considering the influence of simultaneous masking based on the spectral envelope obtained from the linear prediction coefficient. Masking correction means, and a correction amount of the minimum audible value of the hearing considering the influence of the temporal masking based on the temporal change of the power of the input signal over a plurality of frames and the spectral envelope obtained from the linear prediction coefficient. Based on the correction amount obtained by the continuous masking correction means, the simultaneous masking correction means, and the continuous masking correction means, And a masking threshold value determining means for determining a masking threshold value, and a spectrum envelope obtained from the linear prediction coefficient is divided by the masking threshold value, and the power of the spectrum envelope divided by the masking threshold value is changed depending on a plurality of frames. An inter-frame bit allocation control means for adaptively allocating bits to each frame, a codebook having a representative vector code for vector quantizing the orthogonal transform coefficient, and the orthogonal code using the codebook. A voice code comprising: a spectrum parameter quantizing means for determining a code of a representative vector and gain information optimum for a transform coefficient, and quantizing the gain information according to the bit allocation amount. Device. 5. A linear prediction analysis unit for linearly predicting and analyzing a frame-by-frame input signal to obtain a linear prediction coefficient and a prediction residual, and an orthogonal transform for orthogonally transforming the prediction residual into a frequency component to obtain an orthogonal transform coefficient. A conversion means; a power calculation means for obtaining the power of the input signal in frame units; and a simultaneous calculation of a correction amount of the minimum auditory audible value considering the influence of simultaneous masking based on the spectral envelope obtained from the linear prediction coefficient. Masking correction means, and a correction amount of the minimum audible value of the hearing considering the influence of the temporal masking based on the temporal change of the power of the input signal over a plurality of frames and the spectral envelope obtained from the linear prediction coefficient. Based on the correction amount obtained by the continuous masking correction means, the simultaneous masking correction means, and the continuous masking correction means, And a masking threshold value determining means for determining a masking threshold value, and a spectrum envelope obtained from the linear prediction coefficient is divided by the masking threshold value, and the power of the spectrum envelope divided by the masking threshold value is changed depending on a plurality of frames. And an inter-frame bit allocation control means for adaptively allocating bits to each frame, and a code of a representative vector for vector quantizing the orthogonal transform coefficient. And a plurality of codebooks having different required bit amounts corresponding to the codebook size, and a codebook having a codebook size corresponding to the bit allocation amount is selected from the plurality of codebooks. ,
A speech coding apparatus comprising: a spectrum parameter quantizing means for determining a code of a representative vector and gain information optimum for the orthogonal transform coefficient using the selected codebook. 6. The speech coding apparatus according to claim 4, wherein the spectrum parameter quantizing unit determines the code and gain information of the representative vector on a perceptually weighted scale with respect to a frequency axis. A speech coding apparatus characterized in that 7. Band dividing means for dividing a frequency band of an input signal into a first frequency band and a second frequency band lower than the first frequency band, and the first and second frequency bands, Band bit allocation control means for respectively allocating an amount of bits according to the power ratio of signals in each frequency band, and a signal for the first frequency band is input as an input signal, and the bit allocation control means for the first bit allocation control means. A first speech encoding module comprising the speech encoding device according to claim 3, which allocates bits assigned to the frequency band of, and a signal of the second frequency band is input as an input signal. 6. The audio encoding device according to claim 4, wherein the inter-frame bit allocation control means allocates the bits allocated to the second frequency band. And a second speech coding module consisting of a unit.