JPH0473700A - Sound encoding system - Google Patents

Abstract

PURPOSE:To obtain satisfactory characteristic even when code book size is reduced in a low bit rate by correcting a first code book based on a signal selected from a second code book. CONSTITUTION:An adder 290 adds the predictive sound source signal of an adaptive code book 210, the output sound source signal of a first code book retrieval circuit 230, and that of a second code book retrieval circuit 270, and outputs a result to a synthetic filter 281. The synthetic filter 281 inputs the output of the adder 290, and finds one frame of synthesized voice, and finds a response signal series for another one frame by inputting a series of 0 to the filter, and outputs the response signal series of one frame to a subtractor 190. A multiplexer 260 issues output by combining the output encode series of an LSP quantization circuit 140, the adaptive code book 210, the first code book retrieval circuit 230, the second code book retrieval circuit 270, a gain quantizer 286. Thereby, it is possible to obtain the satisfactory characteristic even when the code book size is reduced in the low bit rate.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Field of Application] The present invention provides a method for converting audio signals to low bit rates, particularly from 8 to 4
．． The present invention relates to a speech encoding method for high-quality encoding at approximately 8 kb/s.

(Prior art) As a method for encoding an audio signal at a low bit rate of about 8 to 4.8 kb/s, for example, M, 5 chroe
der and B, “Code-e” by Mr. Ata1
xcited 1inearpredict:on:
lIigh quality 5peech
at very lowbit rates'' (P
roc, TCASSP, pp, 937-940.1
CELP (Code Excited LPCCodi), which is described in a paper titled (Reference 1)
ng) is known. In this method, on the transmitting side, spectral parameters representing the spectral characteristics of the audio signal are extracted from the audio signal every frame (for example, 2 ms),
The frame is further divided into small interval subframes (e.g. 5 ms
), extract pitch parameters from an adaptive codebook that represents long-term correlation (pitch correlation) based on past sound source signals for each subframe, and use the pitch parameters to predict the subframe audio signal over the long term. One type of noise is synthesized using a signal selected from a codebook consisting of predetermined types of noise signals for the predicted residual signal, and the speech signal is synthesized to minimize the error power. Select the signal and calculate the optimal gain. Then, the index and gain representing the type of the selected noise signal, as well as the spectrum parameter and pitch parameter are transmitted.

[Problems to be Solved by the Invention] In the conventional method of Document I mentioned above, in order to obtain high sound quality, it is generally necessary to make the bit size of the codebook composed of noise signals extremely large, to 10 bits or more. Therefore, there is a problem in that a huge amount of calculation is required to search the codebook and find the optimal noise signal (codeword). Furthermore, since the codebook is basically composed of noise signals, there is a problem in that the sound quality of the reproduced sound reproduced by the sound source signal selected from the codebook is accompanied by a noise region. Furthermore, when the size of the codebook is reduced in order to reduce the bit rate, there is a problem in that the sound quality deteriorates rapidly.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems and to provide a speech encoding method that achieves good sound quality at approximately 8 to 4.8 kb/s with a relatively small amount of calculation and memory.

[Means for Solving the Problems] The first invention divides an input discrete audio signal into frames of a predetermined time length, determines and outputs spectral parameters representing the spectral envelope of the audio signal. ,
The frame is divided into small sections of a predetermined time length, a pitch parameter is determined so that the signal reproduced based on the past sound source signal is close to the sound signal, and the sound source signal of the sound signal is In a speech encoding method expressed by a linear combination of a signal selected from a codebook and a signal selected from a second codebook, the signal selected from the second codebook is encoded in the first codebook. It is characterized by correcting.

A second invention divides an input discrete audio signal into frames of a predetermined time length, calculates and outputs a spectral parameter representing a spectral envelope of the audio signal,
The frame is divided into small sections of a predetermined time length, a pitch parameter is determined so that the signal reproduced based on the past sound source signal is close to the sound signal, and the sound source signal of the sound signal is In a speech encoding method expressed by a linear combination of a signal selected from a codebook and a signal selected from a second codebook, the second codebook is generated based on the signal selected from the first codebook. It is characterized by correction.

A third invention divides an input discrete audio signal into frames of a predetermined time length, determines and outputs a spectral parameter representing the spectral envelope of the audio signal,
The frame is divided into small sections of a predetermined time length, pitch parameters are determined for each of the small sections to obtain a pitch predicted sound source signal, and the voice signal is calculated using the pitch predicted sound source signal and the signal selected from the codebook. In the speech encoding method representing a sound source signal, the codebook is modified based on the pitch predicted sound source signal, or the predicted sound source signal is modified using a signal selected from the codebook. .

[Effect] 3 shows the operation of the audio encoding method according to the present invention.

In the first invention, a sound source signal is obtained by minimizing the following equation for each subframe obtained by dividing a frame.

E=Σ [(x (n)-βv (n-M) *
h (n)-d (n)*h (n)l *w
(n)]... (1) Here, β1M is the pitch parameter of the pitch prediction (adaptive codebook) based on long-term correlation, that is, the gain and delay, and v (n) is the past sound source signal.
h(n) indicates the impulse response of the synthesis filter configured by the spectral parameters, and w(n) indicates the impulse response of the perceptually weighted filter. The symbol * indicates a convolution operation. Note that the above-mentioned document 1 can be referred to for details of w (n).

In addition, d(n) indicates the sound source signal represented by the codebook, and as shown in the following equation, the codevector c,(n) selected from the first codebook and the code vector selected from the second codebook are It is expressed as a linear combination with code vector c-(n).

d (n) =ΣTtCt (n) =T+C+i (n) +rz
cz; (n)・・・・(2) Here, Tt+7'z is the selected code word C1
(n), C2, and the gain of (n) are shown. Therefore, in the present invention, since the sound source signal is represented by decomposing two types of codebooks, each codebook may have 1/2 the number of bits of the entire codebook.

For example, if the number of bits of the entire codebook is 10 bits, then the twelfth codebook may have, for example, 5 pins each, which can significantly reduce the amount of calculation for codebook search.

When using a noise codebook as in the above-mentioned document 1 as each codebook, if it is divided as shown in equation (2), the characteristics will be worse than the codebook for 10 bits, and the overall performance will be equivalent to 7 to 8 bits. I can only give out.

Therefore, in order to obtain high performance, the first codebook is configured by learning in advance using training data. As a method of constructing a codebook through learning, for example, “AnAlgorit” by Linde et al.
hm for Vector Quantizatio
The paper titled “De-sign” (IEEE Tra
ns, C0M-28, I) 9.84-951980) (Reference 2), etc. are known.

Squared distance (Euclidean distance) is normally used as a distance measure during learning, but in the present invention, the distance measure is used for perceptual weighting according to the following equation, which has better performance than squared distance.

=Σ [(t, (n) −C1(n)* h (n)
)*W (n))”
(3) Here, tJ(n) is the j-th training data, and c,(n) is the codeword of cluster 1.

The centroid (representative codeword) of cluster l is obtained by minimizing equation (4) or equation (5) using training data within cluster l.

1 Σ Σ [(tit (n) −3cL (n) *
h (n) 1*w (n) ) 2(4) =Σ Σ [(tjL(n) g-set(n)*
h＊w (n) ) ” (
5) In equation (5), g represents the optimal gain.

Next, in order to relieve the training data dependence caused by the first codebook, the second codebook is applied to the remaining signal after subtracting the contribution of the first codebook from the sound source signal. A codebook obtained by learning using the method described above, a codebook consisting of a noise signal or random number signal with predetermined statistical characteristics such as the Gaussian noise signal described in Document 1, or a code having other characteristics. Use books. Note that the characteristics can be further improved by performing learning using a certain distance measure with respect to the noise codebook. For details, see “Transform” by T0 Moriya et al.
Coding of 5peach using a
Weighted Vector Quantizer
,” (IEEEJ, Se1.^reas
Common, pp. 425-43L 19B8) (Reference 3).

The first invention is characterized in that the first codevector is modified using the codevector selected by the second codebook. This is done in order to adapt the first codebook to the short-time characteristics of the input signal and obtain better characteristics with a smaller codebook size. The first codebook is modified according to the following formula.

C'lj (n) ”: (+j(n) +sign (r+) ・sig
n (rz) ・B°cz+(n)
(6) Here, B is a positive minute quantity that determines convergence. Further, sign(T) represents the sign of T. (
6) Modify the first codebook on both the transmitting side and the receiving side according to the formula.

Furthermore, in order to make the signal more resistant to transmission path errors and to prevent the influence of erroneous corrections, the following equation can be used instead of (6) 2.

cl, J(n) −(1−δ)−c,,(n)+A・δ′cz;(n)・
... (7) Here, δ is a positive minute amount (for example, 104 to 10
-'), A is a convergence coefficient determined by the following formula.

A-72/r+
(8) Here, if the error between the input vector and the vector selected in the first stage becomes smaller, the ratio Tz/T+ of the gains in the first stage and the second stage becomes smaller, making it difficult to proceed with the correction.

Furthermore, simplified equations (9) and (10) can also be used.

C', j (n) = (1-δ)・C1J(n) 10A・δ (9)
C'lj (n) -(1-δ)・c+J(n) +sign (r+) ・sign (rz) ・
δ (10) Next, in the second invention, the second codebook is modified using the code vector selected from the first coat hook. For the modification, equation (11) or equation (12) is used.

C'zi(n) = siguro (T+) ・ sign (rz)
・ B −C+j (n)+ Czr (n
) (11) C'zi
(n) =A・δ−C1j (n)+ (l−δ) −Chi
(rl)...(12) Furthermore, simplified equations (13) and (14) can also be used.

C'z((n) −”δ+(1−δ) ・Chi (
n)・・・・(13) +2゜(n) =sign (r+) °sign (T2) 0
δ+ (1-δ) ・Czt (n)
(14) Furthermore, in the third invention, the sound source signal of the audio signal is represented by an adaptive codebook and a coat hook as in the above-mentioned document 1. The code vector selected in the codebook is corrected using the pitch predicted excitation signal obtained based on the past excitation signal using the adaptive codebook. The modification follows equation (15) or equation (16).

C', (n) = sign (β) ・sign (7+) B-v
(n-M)±Cj'(n)
(15) C', (n) -A・δ-v (n-M) + (1-δ)-Cj(n
)...(16) Here, v(n-M) is the pitch predicted excitation signal obtained using the adaptive codebook. Further, c=(n) is the j-th code vector selected from the codebook.

Furthermore, simplified equations (17) and (18) can also be used.

C', (n) =A・δ+(1−δ)・cJ(n) (1
7) C'J(n) -sign (β) ・sign(r+) Hδ+
(1-δ)・C5(n) (18
) Also, the code vector C selected from the codebook
4(n) can also be used to modify the pinch prediction source signal.

[Embodiment] FIG. 1 is a block diagram showing a speech encoding device implementing the speech encoding method according to the first invention.

On the transmitting side, an audio signal is input from the input terminal 110, and 1
The audio signal for one frame (for example, 20 m5) is stored in the buffer memory 120.

The LPG calculation circuit 130 performs well-known LPG analysis on the frame audio signal to calculate LSP parameters of predetermined orders as parameters representing the spectral characteristics of the frame audio signal. Regarding this specific calculation method, reference can be made to the above-mentioned document 1.

Next, the LSP quantization circuit 140 quantizes the LSP parameters with a predetermined number of quantization bits, and the obtained code lk
is output to the multiplexer 260, and is decoded and further converted into linear prediction coefficients a;' (i=1 to L), and the weighting is performed by the circuit 200. Impulse response calculation circuit 1701 is output to synthesis filter 281. Regarding the method of encoding LSP parameters and converting LSP parameters into linear prediction coefficients, see Quantizer design in LSP 5 by Mr. Sugamura et al.
IEEE J, Sel.

Areas Commun,, pp, 432-440
．． 1988) (Reference 4).

The subframe division circuit 150 divides the frame audio signal into subframes. For example, the frame length is 2
0m5, and the subframe length is 5ms.

The weighting circuit 200 performs well-known perceptual weighting on the subtracted signal. For details of the perceptual weighting function, refer to the above-mentioned document 1.

The subtracter 190 subtracts the output of the synthesis filter 281 from the human signal divided into subframes and outputs the result.

The adaptive codebook 210 inputs the human input signal v (n) of the synthesis filter 281 via the delay circuit 206, and further inputs the weighted impulse response hw (n) from the impulse response output circuit 170 and the weighted signal from the circuit 200. Next, a signal is input, pinch prediction is performed based on long-term correlation, and delay M and gain β are calculated as pitch parameters. In the following explanation, the prediction order of the adaptive codebook is assumed to be 1, but it can also be set to a higher order than 2. The calculation method for delay M and gain β in the first-order adaptive codebook is as follows:
Kleijin"Improved 5peech q
quality and efficiency
or quantization in 5ELP” (Proc.

ICASSP, pp, 155-158.1988)
(Reference 5) etc. Further obtained gain β
is quantized and decoded by a gain quantizer using a predetermined number of quantization bits to obtain a gain β'. Using this, a prediction signal 9w(n) is calculated by the following equation and output to the subtracter 205. It also outputs the delay M to the multiplexer 260.

xw(n)-β'-v (n-M) * h, (n
) (19) In the above equation, v(n-M) is a past sound source signal and is an input signal of the synthesis filter 281. h and (n) are the weighted impulse responses obtained by the impulse response calculation circuit 170.

The delay circuit 206 delays the synthesis filter input signal v (n) by one subframe and outputs the delayed signal to the adaptive codebook 210 .

The subtracter 205 subtracts the output of the adaptive codebook 210 from the output signal of the weighting circuit 200 using the following equation, and outputs the residual signal e,(n) to the first codebook search circuit 230.

el, 1(n)=x, 1(n)-x, (n)
(20) The impulse response calculation circuit 170 calculates the impulse response hl,(n) of the auditory weighted synthesis filter for a predetermined number of samples. For a specific calculation method, reference can be made to the above-mentioned document 1, etc.

The first codebook search circuit 230 uses the first codebook 235 to search for the optimal codeword C1j(n). As described in the section on effects, the first codebook is learned in advance using a training signal. For a method of searching for the optimal code vector C+=(n), reference can be made to Japanese Patent Application No. 2-42956 (Reference 6). Then, find the optimal gain γ1 and use this and C1j(n)
The weighting is performed using the method of the above-mentioned document 6 using the reproduced signal yw
(n) is found and output.

The subtracter 255 subtracts y,(n) from ew(n) and outputs the result to the second codebook search circuit 270.

Second codebook search circuit 270 calculates an optimal codeword from second codebook 275. The configuration of the second codebook search circuit can be basically the same as the configuration of the first codebook search circuit. Further, as a codeword search method, the same method as the search for the first codebook 235 can be used. As described in the operation section, the second codebook is constructed using a codebook consisting of a random number sequence in order to maintain the high efficiency of the learning codebook and relieve the training data dependence. The above-mentioned document 1 can be referred to for a method of constructing a codebook consisting of a random number sequence.

In addition, in order to reduce the amount of calculation for codebook search, a superimposed type (0νerla) is used as the second codebook 275.
p) A random number codebook can be used. Regarding the construction method of the superimposed random number codebook and the codeword search method, reference can be made to the above-mentioned document 5 and the like.

Further, like the first codebook, it can be configured by learning in advance.

The gain quantizer 286 calculates the gain γ by using the gain code bank 287 created in advance using equations (12) and (13) by learning according to the method described in the operation section.

Vector quantize γ2. For details, refer to the above-mentioned document 6.

The correction circuit 280 performs (6) to (10) described in the operation section.
The code vector C1j(n) selected by the first codebook search circuit 230 is modified using the formula.

The adder 290 adds the predicted excitation signal of the adaptive codebook 210, the output excitation signal of the first codebook search circuit 230, and the output excitation signal of the second codebook search circuit 270, and sends the result to the synthesis filter 281. Output.

The synthesis filter 281 outputs the output v (n) of the adder 290
is input, calculate one frame of synthesized speech using the formula below, and input a series of 0s for one more frame into the filter to obtain a response signal sequence. 190.

(0<η<1) (21) However, the multiplexer 260 is connected to the LSP quantization circuit 140. Adaptive Code Zinc 210. First codebook search circuit 2
30. Second codebook search circuit 270. The output code sequences of the gain quantizer 286 are combined and output.

This concludes the detailed description of the first invention.

FIG. 2 is a block diagram showing a speech encoding device implementing the speech encoding method according to the second invention.

In the figure, components numbered the same as in Figure 1 are as follows:
Since the components perform the same operations as those shown in FIG. 1, their explanation will be omitted.

The correction circuit 380 is the first codebook search circuit 230
Using the code vector C1j(n) selected in the second codebook search circuit 270, based on the equations (11) to (14) described in the operation section, the code vector Cz;(n) Make corrections.

This concludes the explanation of the second invention.

FIG. 3 is a block diagram showing a speech encoding device implementing the speech encoding method according to the third invention.

In the figure, components numbered the same as in Figure 1 are as follows:
Since the components perform the same operations as those shown in FIG. 1, their explanation will be omitted.

The sound source codebook search circuit 430 performs the same operation as the first codebook search circuit 230 and selects the optimal sound source signal from the sound source codebook 435.

The correction circuit 480 uses the pinch prediction excitation signal v(n-M) obtained using the adaptive codebook 210 to calculate (15)
~ Using equation (18), the sound source codebook search circuit 43
The sound source signal C4, (n) selected by 0 is corrected.

This completes the detailed description of the third invention.

In addition to the method described in the embodiment, the following method can also be used as a correction method in the correction circuit. For example, taking the first invention as an example, C'lj (n) -(1-δ) ・C+j (n) +sign・l
T2/ 'r・δ・Cz+ (n)
(23) Here, sign indicates a positive or negative sign. The code is selected to minimize the following equation.

E−Σ [et, + (n)”−TIC'la (n)
*h,,(n)] 2 (24)
Here, e,1(n) is the output signal of the subtracter 205.

In order to minimize the above equation, it is sufficient to partially differentiate the above equation with respect to T1 and minimize the below equation, which is set as O.

-r, X ew (n)fc'+; (n)*hw(n
)l k ”/Σ (C'+i (n) *hw (
n)) 2 (2,1)) Therefore, calculate equation (25) for both positive and negative signs of (23)2, and set the sign of the one where equation (25) takes a larger value to 1. Transmit with focus.

The second and third inventions can also have a configuration similar to that described above.

Further, in the above-described embodiment, the gain of the adaptive coat 7 and the first . Although we did not perform simultaneous optimization on the gain of the second codebook, or the gain of the adaptive codebook and the gain of the sound source codebook, we did not perform simultaneous optimization on the gain of the adaptive codebook, the first codebook, and the second codebook. Perform optimization and further improve characteristics. This simultaneous optimization performs gain optimization only when searching for the code vector of the first codebook in order to reduce the amount of calculation when calculating the code vector of the first codebook. It is also possible to adopt a configuration in which the search is not performed when searching the second codebook.

In addition, in order to further reduce the amount of calculation, the code edge of the code book!・When searching for the code, the gains of the adaptive codebook and the first codebook are simultaneously optimized without optimizing the gains, and the gains of the adaptive codebook and the first and second codebooks are simultaneously optimized. Optimizing configurations may also be used. For details, refer to the above-mentioned document 5 and the like.

In addition, in order to further reduce the amount of calculation, after the code vector of the 1.2nd codebook is selected, the gain β of the adaptive codebook and the gain T++Tz of the 12th codebook are simultaneously calculated. It is also possible to configure a configuration for optimization. For details, refer to the above-mentioned document 6.

In addition to the method of the embodiment, other well-known methods can be used as the first codebook search method. For example, the method described in the above-mentioned document 1 or the orthogonal transformation CI(k) of each codeword C+;(n) of the codebook is determined and stored in advance, and the weighting is done by impulse response for each subframe. The orthogonal transformation H, (k) of (n) and the residual signal e, (n
It is also possible to obtain orthogonal transformation E1.1(k) of ) by a predetermined number of points and search on the frequency axis. For details, refer to the above-mentioned document 5 and the like.

Furthermore, as a search method for the second codebook, in addition to the method of the above embodiment, the method shown above, the method described in the above-mentioned document 6, and other well-known good methods can be used.

In addition, as a method for configuring the second codebook, in addition to the method described in the above embodiment, for example, a huge number of random number sequences can be prepared in advance as a codebook, and these can be used to search for a random number sequence for training data. It is also possible to construct a second codebook by registering the codewords as codewords in descending order of frequency of selection. Note that this configuration method can also be applied to the configuration of the first codebook.

Furthermore, in the above embodiment, the gain of the adaptive codebook and the first . Although the gains of the second codebook are vector-quantized separately, it is also possible to adopt a configuration in which the three types of gains β, TT2 or β, T1 are vector-quantized together. For details, see the above-mentioned document 5 and 1. “Vector sum excited 1ine” by Mr. Gerson et al.
ar prediction” (VSELP) 5p
1 imperative entitled ``eech codingat 8kbp/s'' (Proc, ICASSP, pp.

461-464.1990) (Reference 7).

Further, in the above embodiment, the order of the adaptive code book is 1.
Although the following is used, it is also possible to set it to a higher order than second order. Further, it is also possible to set the delay to a decimal value instead of an integer value while keeping the order as one. For details about these, see P.
'Pi tch predic by Kroon et al.
tors with high temporal r
A paper titled “Proc.
ASSP, pp, 661-664.1990) (Reference 8), etc. can be referred to. Although the above method improves the characteristics, the amount of information necessary for transmitting gain or delay increases slightly.

Furthermore, in the above embodiment, parameters and LSP parameters were encoded as spectral parameters, and LPG analysis was used as an analysis method thereof. However, as spectral parameters, other well-known parameters such as LPC cepstrum, cepstrum, improved cepnutrum, Generalized cepstrum, merkebstrum, etc. can also be used. Furthermore, it is possible to use the optimal analysis method for each parameter.

In addition, the LPC coefficients obtained for each frame are interpolated on LSP and linear prediction coefficients for each subframe, and the interpolated coefficients are used to create an adaptive codebook. A configuration may also be adopted in which a second codebook is searched. With such a configuration, the sound quality is further improved.

Further, the LSP coefficients can be encoded more efficiently by vector quantization or hectorose scalar quantization using a well-known method.

For the method of hectose scalar quantization, reference can be made to, for example, the above-mentioned document 3.

Furthermore, on the receiving side, an adaptive post filter that operates on at least one of pitch and spectrum envelope may be added in order to make the quantization noise easier to hear by shaping it. Regarding the configuration of the adaptive post filter, for example, "A C1ass" by Kroon et al.
of analysis-by-synthesis
Predictive Coders for
) figure [1uality 5peechCo
din(Hat Rates betiyeen 4
．． 8 and 16kb/s,” (JEEEJS
AC, vol, 6.2.353-363.1988)
(Reference 9) etc. can be referred to.

〔Effect of the invention〕

As described above, according to the present invention, the first or second
The codevector selected by the codebook is modified based on the codevector selected by the second or first codebook, or the codevector selected by the adaptive codebook is modified based on the pitch predicted excitation signal selected by the adaptive codebook. Since the code vector selected by the sound source codebook is modified or the pitch predicted sound source signal is modified by the code vector, it is possible to adapt the characteristics of the codebook to the characteristics of the human signal, This method has the great effect of providing better characteristics than the conventional method even if the codebook size is reduced at low bit rates.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a speech encoding device implementing the speech encoding method according to the first invention, FIG. 2 is a block diagram showing a speech encoding device implementing the speech encoding method according to the second invention, FIG. 3 is a block diagram showing a speech encoding device implementing the speech encoding method according to the third invention. 110 ...Buffer memory 130 ...LPC calculation circuit 140 ...Quantization circuit 150 ...Subframe division circuit 170
... Impulse response calculation circuit 190, 205.2
55...Subtractor 200...Weighting circuit 206...Delay circuit 210...Adaptive codebook 230...
...First codebook search circuit 235...
-First codebook 281...Synthesis filter 270...Second codebook search circuit 27
5...Second code bunk 380, 480
... Correction circuit ... Gain quantizer ... ... Gain codebook ... Sound source codebook search circuit ... Sound source codebook

Claims (3)

[Claims] (1) Divide the input discrete audio signal into frames of a predetermined time length, obtain and output spectral parameters representing the spectral envelope of the audio signal, and divide the frames into frames of a predetermined time length. The pitch parameter is determined so that the signal reproduced based on the past sound source signal is close to the sound signal, and the sound source signal of the sound signal is divided into intervals between the signal selected from the first codebook and the second codebook. In a speech encoding method expressed by linear combination with a signal selected from a codebook, the speech encoding method is characterized in that the first codebook is modified based on the signal selected from the second codebook. . (2) Divide the input discrete audio signal into frames of a predetermined time length, obtain and output spectral parameters representing the spectral envelope of the audio signal, and divide the frames into frames of a predetermined time length. The pitch parameter is determined so that the signal reproduced based on the past sound source signal is close to the sound signal, and the sound source signal of the sound signal is divided into intervals between the signal selected from the first codebook and the second codebook. A speech encoding method represented by linear combination with a signal selected from a codebook, characterized in that the second codebook is modified based on the signal selected from the first codebook. . (3) Divide the input discrete audio signal into frames of a predetermined time length, obtain and output spectral parameters representing the spectral envelope of the audio signal, and divide the frames into frames of a predetermined time length. A speech encoding method in which the speech signal is divided into sections, a pitch parameter is obtained for each of the small sections to obtain a pitch predicted excitation signal, and the speech source signal of the speech signal is expressed by the pitch predicted excitation signal and a signal selected from a codebook. A speech encoding method characterized in that the codebook is modified based on the pitch predicted excitation signal, or the predicted excitation signal is modified using a signal selected from the codebook.