JPH034300A - Voice encoding and decoding system - Google Patents

Abstract

PURPOSE:To reproduce a voice excellently with a small arithmetic quantity irrelevantly to a bit rate by dividing a frame into subframes corresponding to a pitch period and using multipulses which are found by pitch interpolation and multipulses which are found on the whole frame without pitch prediction. CONSTITUTION:The voice signal from an input terminal 100 is divided into specific frames and an LPC and a pitch analysis part 150 finds an LPC coefficient of specific degree as a spectrum parameter representing a spectrum envelope from the voice signal of a frame. A pitch interpolation multipulse calculator 250 divides the frame into subframes by using a pitch period, finds the amplitude and position of multipulses in one specific section, and finds gain correction coefficients and phase correction coefficients of other subframes. Then a sound source signal is restored by specific arithmetic operation through a subtracter 260 and a multipulse calculation part 270 and then processed by composition. Consequently, the excellent voice can be reproduced with a small arithmetic quantity irrelevantly to the bit rate.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an audio encoding and decoding method for efficiently encoding and decoding audio signals at low bit rates.

(Prior Art) As a method for transmitting audio signals at a low bit rate, for example, about 16 Kb/s or less, a multipulse encoding method is known. These represent the sound source signal as a combination of multiple pulses (multi-pulse), represent the characteristics of the vocal tract with a digital filter, and transmit the information on the sound source pulse and the filter coefficients after determining them for each fixed time interval (frame). There is.

For details of this method, see for example Araseki,
“Mu” by Ozawa, Ono, Ochiai
lti-pulse Excited 5peech
Coder Ba5ed on Maximum Cr
osscorrelation 5earch A
lgorithm”.

(GLOBECOM 83. IEEE Global
Telecommunication.

It is described in lecture number 23.3.1983X document 1). This method can output a good audio signal after decoding by separately expressing the vocal tract information and the sound source signal, and by using a combination of multiple pulse trains (multipulse) as a means of expressing the sound source signal. . The basic idea of finding a pulse train representing a sound source signal will be explained using FIG. 5. An audio signal divided into frames is manually inputted from an input terminal 900 in the figure. Spectral parameters determined from the audio signal of the current frame are input to the synthesis filter 920. An initial multipulse is generated in the sound source calculation circuit 910 and inputted to the synthesis filter 920 to obtain a synthesized speech waveform as an output. A subtracter 940 subtracts the synthesized speech waveform from the human input signal. This result is weighted by circuit 9.
50 and weighting in the current frame obtains the error power. Then, the amplitude and position of a specified number of multipulses are determined in the sound source calculation circuit 910 so that this weighting minimizes the error power.

(Problem to be solved by the invention) However, with this conventional method, the sound quality was good when the bit rate was high enough and the number of sound source pulses was sufficient, but as the bit rate was lowered, the sound quality deteriorated. I agree with the problem.

In order to improve this problem, a pitch prediction multi-pulse method has been proposed that utilizes the pitch-wise quasi-periodicity (pitch correlation) of a multi-pulse sound source. The details of this method are detailed in, for example, Japanese Patent Application No. 139022/1982 (Document 2), so the explanation will be omitted here.

However, although the pitch-wise quasi-periodicity of a multipulse sound source is considered to be large for large-amplitude pulses, such periodicity does not exist for all pulses, and small-amplitude pulses have pitch-wise periodicity. is considered to be small. In the pitch prediction multi-pulse method of Document 2, all pulses are determined by pitch prediction assuming periodicity for each pitch for a predetermined number of pulses within a frame, so all pulses are determined by pitch prediction. For pulses, there is a problem in that pitch prediction actually worsens the characteristics. This is particularly noticeable in transition sections and transitional parts between vowels, and there is a problem in that the sound quality deteriorates in such parts.

Furthermore, in the method of Document 2, since pitch information is included in the impulse response, a very long impulse response (for example, 20 m5ec or more) is required, and all pulses of a predetermined number are determined by pitch prediction. Since we are searching for a pulse, the amount of calculation required to search for a pulse is extremely large.
Even with current LSI technology, it has been difficult to realize a compact device.

The purpose of the present invention is to provide an audio encoding/decoding method that can reproduce better audio than before even when the bit rate is high or lower, and that can be realized with a small amount of calculation. be.

(Means for Solving the Problem) The audio encoding/decoding method of the present invention inputs a discrete audio signal on the transmitting side, and extracts a spectral parameter representing a spectral envelope and a pitch representing a pitch period from the audio signal for each frame. The audio signal of the frame is divided into small sections according to the pitch parameter, and the audio signal of one section of the small sections is extracted using the pitch parameter and the spectrum parameter. 1 multi-pulse is obtained, coefficients for correcting the multi-pulse are obtained in other sections, and the signal obtained by removing the multi-pulse and the coefficient from the audio signal is obtained using spectral parameters. A second multi-pulse is obtained, and on the receiving side, a sound source signal is restored using the first multi-pulse, the pitch parameter, the coefficient, and the second multi-pulse, and further configured using the spectral parameter. It is characterized by driving a synthesis filter to obtain a synthesized speech signal.

Further, in the audio encoding method according to the present invention, a discrete audio signal is input on the transmitting side, and a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch period are extracted from the audio signal for each frame, and the audio signal of the frame is is divided into small sections according to the pitch parameter, a first multi-pulse is obtained as a sound source signal of the audio signal using the pitch parameter and the spectrum parameter in one section among the small sections, and the first multi-pulse is obtained in the other sections. Then, a coefficient for correcting the multi-pulse is obtained, and a second multi-pulse is obtained by using the spectral parameter for the signal obtained by removing the multi-pulse and the signal obtained by the coefficient from the audio signal. The receiver side uses a multi-pulse sound source selected from a codebook consisting of predetermined types of noise signals to reduce the error power between the speech signal and the synthesized signal. 1 multi-pulse, the pitch parameter, the coefficient and the second multi-pulse, or the selected noise signal is used to restore the sound source signal and the spectral parameter is used to reconstruct the sound source signal. The synthesized speech signal is obtained by driving a synthesis filter according to the sound source signal.

(Operation) The audio encoding/decoding method according to the first invention converts the sound source signal of an audio signal in a frame section (for example, 20 m5) into a multi-pulse ( first multi-pulse);
It is characterized by representing the entire frame using a multi-pulse (second multi-pulse) obtained without pitch prediction. The calculation of the first multi-pulse is performed as follows. In order to utilize the monoperiodic nature of each pitch of a multipulse sound source very efficiently and to greatly reduce the amount of calculation, a frame is divided in advance into small sections (subframes) according to the pitch period, and each of the subframes is Multipulses are obtained only for one subframe (representative section). For other subframes, find a correction coefficient that corrects the gain and phase of the multipulse found in the representative section, and use this coefficient to correct the gain and phase of the multipulse in the representative section in other subframes. to generate a pulse and restore the pulse for the entire frame. Then, after reproducing a signal in frames using the pulses and subtracting the signals from the audio signal, a multipulse (second multipulse) is obtained in the frame using the same method as in Document 1.

The basic processing of this method will be explained below using FIG. FIG. 3 is a block diagram showing the operation of the present invention. An audio signal is input from an input terminal 100, and the audio signal is divided into frames of a predetermined time length (for example, 20 m5). The LPG and pitch analysis unit 150 converts the LPG coefficients of a predetermined order into well-known LPC signals from the frame audio signal as spectral parameters representing the spectral envelope.
Stop for analysis. In addition to the linear prediction coefficient a used here, LPG coefficients include LSP, formant, and LPC.
Other good parameters such as cepstrum can also be used. Furthermore, analysis methods other than LPC, such as cepstrum, PSE, and ARMA methods, can also be used.

The following explanation assumes that linear prediction coefficients are used.

Further, 150 calculates a pitch period M as a pitch parameter from the audio of the frame. The well-known autocorrelation method can be used for this.

The operations of the pitch interpolation multipulse calculation section 250 and the multipulse calculation section 270 will be explained with reference to FIG. Fourth
Figure (a) represents the audio signal of a frame. Here, as an example, the frame length is 20 m5. The pitch interpolation multipulse calculation unit 250 first divides the frame into small sections (subframes) using the pitch period M, as shown in (b).
Divide into. Here, the length of the subframe is the same as the pitch period M.

Next, using the same method as in Document 1, the impulse response h(n
), the autocorrelation function ``hh(m)'', the cross-correlation function ohx between the perceptually weighted speech signal and the impulse response h(n)
Find (m). Next, a predetermined number K (here, 4 amplitude g4 and position m of the multi-pulse (first multi-pulse).

seek. Here, reference can be made to the above-mentioned document 1 for how to obtain the multi-pulse. FIG. 4(C) shows the obtained multipulse. Next, in subframes other than the representative section, gain correction coefficients and phase correction coefficients for generating pulses by correcting the gain and phase of the multi-pulse obtained in the representative section are determined. The gain correction coefficient C1 and phase correction positive coefficient d in the j-th subframe within the frame are determined to minimize the error power according to the following equation.

E=E[(xt(n) −y(n))*w(n)]
(1) Here x, (n), s,
(n) shows the synthesized speech obtained by correcting the sound 7jg signal, multi-pulse gain, and phase in the j-th subframe. However, S, (, n)
=ciig, g, ・h (n-m, 7L-M-d,)
(L is an integer) (2) where h(n)l is the impulse response of the synthesis filter. By substituting equation (2) into equation (1), partially differentiating it with respect to C0, and setting it as O, C5,d, which minimizes equation (1), can be found. For details, please see the special request
Reference can be made to the specification of No. 3-208201 (Document 3).

In this way, gain correction coefficients and phase correction coefficients are basically obtained for all other subframe sections within the frame. Then, using the multi-pulse of the representative section, the gain correction number, and the phase correction coefficient, the pulse of the entire frame is reproduced as shown in FIG. 4(d). Note that the intra-frame position of the representative section may be determined by searching several subframes, or may be determined in advance. For details of the former method, reference can be made to, for example, the above-mentioned document 3.

Next, a synthesis filter defined by equation (3) is driven using the reproduced pulse v(n) to obtain a reproduced signal x'(n).

x'(n)=v(n)+, fi aix'(ni)
(3) s=1 where a is a linear prediction coefficient.

The subtracter 260 converts the audio signal x(n) to X according to the following equation:
- Subtract (n) to obtain e(n).

e(n) = x(n) −x'(n)
(4) Next, the multipulse calculation unit 270 calculates e(n) using the same method as in Document 1.
A predetermined number Q of multi-pulses are generated within a frame using a cross-correlation function between the perceptually weighted signal (n) and the weighted impulse response of the synthesis filter, and an autocorrelation function of the weighted impulse response. (second multipulse). This is shown in FIG. 4(e). In the diagram, Q
is set as 4.

On the other hand, for an unvoiced frame, the amplitude and position of multipulses are determined for the entire frame.

In addition to the spectral parameters of the synthesis filter, the transmission information on the transmitting side includes, in voiced frames, the spectral parameter a representing the spectral envelope, the pitch M, the amplitude and position of the multipulses in the representative section, the gain correction coefficient, and the phase. These are the correction coefficient, the position within the frame of the representative section, and the amplitude and position of the Q multipulses. Furthermore, in the unvoiced frame, the amplitude and position of the multipulse are transmitted.

In the second invention, the same operation as in the first invention is performed in voiced frames, but in unvoiced frames, a noise signal selected from a codebook of predetermined types of noise signals is used instead of a multi-pulse. It is characterized in that it represents a sound source signal using As the noise signal, for example, a random number having a Gaussian statistical distribution can be used. The length (number of dimensions) of the noise signal in the time direction is shorter than the normal frame (for example, 5 to 10 m5). Furthermore, there are two types of noise signals. The method of selecting the best noise signal for input speech from such a codebook is to use the noise signal to drive a synthesis filter to synthesize speech, find the error power with the original speech, and calculate the error power. Methods of selecting noise signals to be minimized are known. Details of this method can be found, for example, in “Code-
Excited 1inear prediction
(CELP): Highquality 5pe
ech at very low bit rates
” (Proc, ICASSP, pp.
937-940.1985X document 4), etc. can be referred to.

In the unvoiced frame, an index indicating the selected noise signal, a gain, a pitch gain of a pitch recovery filter, a pitch period, and a spectrum parameter of a synthesis filter are transmitted to the receiving side.

(Embodiment) In FIG. 1 showing an embodiment of the first invention, a discrete audio signal x(n) is input from an input terminal 500.

The spectral and pitch parameter calculation circuit 520 calculates the spectral parameter ai of the synthesis filter representing the audio signal spectral envelope of the divided frame section (for example, 20 m5) using the well-known LPC analysis method. Further, the pitch period M is determined by the well-known autocorrelation method.

A quantizer 525 performs quantization on the obtained spectrum parameters and pitch periods.

The quantization method may be scalar quantization as shown in Japanese Patent Application No. 59-272435 (Reference 5), or vector quantization. For a specific method of vector quantization, see, for example, “Vector quantization” by Makhou et al.
in 5peech coding” (Proc, I
You can refer to papers such as EEE, pp. 1551-1558.1985X Reference 6).

The dequantizer 530 dequantizes and outputs the quantized result.

A subtracter 535 subtracts the influence signal from the audio signal of the frame and outputs the result.

Weighting circuit 540 perceptually weights the audio signal using the dequantized spectral parameters. The weighting method of the above-mentioned document 2 is based on the weighting circuit 20.
0 can be referenced.

The impulse response calculation circuit 550 calculates the dequantized spectrum parameters a1. The impulse response h(n) of the synthesis filter subjected to auditory weighting is calculated using .

For a specific method, refer to the impulse response calculation circuit in Document 2.

The autocorrelation function calculation circuit 560 calculates the autocorrelation function "hh(m)" for the impulse response, and outputs it to the sound source pulse calculation circuit 580 and the pulse calculation circuit 586, respectively.The method for calculating the autocorrelation function is described in the above-mentioned document 2. The autocorrelation function calculation circuit 180 of FIG.

A cross-correlation function calculation circuit 570 calculates a cross-correlation function Φ, , (m) between the perceptually weighted signal and the impulse response h(n).

The sound source pulse calculation circuit 580 first divides the frame into subframe sections as shown in FIG. 4(b) using the inversely quantized pitch period M'. Then, for one predetermined subframe section (representative section) (for example, subframe ■ in Fig. 4(b)), Φ, , (m) and R
, , (m) to find the amplitude g of the multi-pulse train (first multi-pulse) at the position mi. Regarding the pulse train calculation method, reference can be made to the sound source pulse calculation circuit in Document 2.

The correction coefficient calculation circuit 583 is shown in the function section (1).

According to equation (2), calculate the gain correction coefficient C1 and the phase correction coefficient d in subframe sections other than the representative section.
” is output.

A quantizer 585 quantizes the amplitude and position of the multi-pulse train and outputs a code. The specific method is in the above document 1.
．． 2 etc. can be referred to. It also quantizes the gain correction coefficient, phase correction coefficient, and position within the frame of the representative section, and outputs a code. For a specific method, reference can be made to the above-mentioned document 3, for example. These outputs are further dequantized and output to a pitch interpolation circuit 605 to restore the pulses of the entire frame as shown in FIG. 4(d).

The restored pulse is passed through a synthesis filter 610 to produce a synthesized speech signal x'(n
) can be found.

The subtracter 615 obtains a residual signal e(n) by subtracting the synthesized speech signal x'(n) from the speech signal x(n) according to equation (4).

The weighting circuit 600 performs perceptual weighting on the residual signal.

A cross-correlation function calculation circuit 603 calculates a cross-correlation function between the output of the weighting circuit 600 and the impulse response h(n).

The pulse calculation circuit 586 uses the cross-correlation function and the autocorrelation function of the impulse response h(n) to find the amplitude and position of a predetermined number of multipulses (second multipulse).

The quantizer 620 quantizes and outputs the amplitude and position of the multi-pulse, and also dequantizes and outputs them to the synthesis filter 625.

A synthesis filter 625 synthesizes and outputs the residual signals.

The adder 627 includes the synthesis filter 625 and the synthesis filter 61.
The reproduced signal of the frame is obtained by adding the outputs of 0, and the influence signal for the next frame is also determined and output. For a specific method of calculating the influence signal, refer to the above-mentioned document 2.

The multiplexer 635 receives the amplitude, position, correction coefficient, and code representing the position of the representative section of the multi-pulse train output from the quantizers 585 and 620, the spectrum parameter output from the parameter quantizer 525, and the code representing the pitch period. Combine and output.

On the other hand, on the receiving side, the demultiplexer 710 outputs the amplitude, position, and correction coefficient of the pitch interpolation multipulse (first multipulse), a code representing the position of the representative section, the amplitude of the multipulse (second multipulse), The code representing the position, the spectrum parameter, and the code representing the pitch period are separated and output.

The first pulse decoder 720 decodes the amplitude and position of the pitch interpolated multi-pulse. A second pulse decoder 725 decodes the amplitude and position of the second multi-pulse. The parameter decoder 750 has the same function as the inverse quantizer 530 on the transmitting side, and has a spectral parameter a''1 and a pitch period M'.
Decode and output.

The pitch interpolation circuit 726 is the pitch interpolation circuit 60 on the transmission side.
Perform the same operation as 5.

The pulse generator 727 generates a sound source signal based on the second multi-pulse for a frame length.

Adder 740 includes pulse generator 727 and pitch interpolation circuit 7
26 output signals are added to obtain a frame driving sound source signal, and the synthesis filter circuit 760 is driven.

The synthesis filter circuit 760 uses the driving sound source signal and the decoded spectral parameters to obtain and output a synthesized speech waveform for each frame.

This concludes the description of one embodiment of the first invention.

FIG. 2 is a block diagram showing an embodiment of the second invention. In the figure, the components numbered the same as in Figure 1 are as follows:
Since the operation is the same as in FIG. 1, the explanation will be omitted.

In the figure, spectrum and pitch parameter calculation circuit 5
22, a spectral parameter a is determined using a well-known LPC analysis, and pitch parameters such as a pitch period M and a pitch gain b are determined using a well-known autocorrelation method.

The quantizer 522 converts the spectral parameter a into PARC
After converting into OR coefficients or LSP coefficients, quantization is performed. Here, PARCOR coefficients are used. Also, the pinch period M and pitch gain b are quantized. Furthermore, these quantized values are decoded to output decoded values a, t, M', and b'.

B.

The codebook 800 stores two types of noise signals (B indicates the number of hits) in advance. For the method of generating the noise signal, refer to the above-mentioned document 4. Of these, one type is output to the convolution circuit 810.

The convolution circuit 810 convolves one type of noise signal c(n) and the impulse response h(n) according to the following equation, and outputs the result to the switch 820.

f(n)=c(n)*h(n)
(5) Here, the symbol * represents a convolution sum.

The switch 820 outputs the output of the impulse response calculation circuit 550 to the correlation function calculation circuit 560 in a voiced frame,
For unvoiced frames, the output of the convolution circuit 810 is output to the autocorrelation function calculation circuit 560. Here, voiced or unvoiced can be determined, for example, when the value of the decoded pinch gain b' exceeds a preset threshold, it is determined that there is voice, and otherwise, it is determined that voice is unvoiced.

The switch 825 outputs the output of the autocorrelation function calculation circuit 560 to the excitation pulse calculation circuit 580 in a voiced frame, and to the signal selection circuit 830 in an unvoiced frame.

The signal selection circuit 830 uses the cross-correlation function Φxh and the autocorrelation function Rhh to calculate the following equation.

G=(ΦXh)/Rhh(6) Calculate equation (6) for all noise signals, and (6)
The noise signal that maximizes the equation is selected, and the index representing the selected noise signal and the gain G obtained from equation (6) are output.

Encoder 840 quantizes gain G using a predetermined number of bits and outputs it to multiplexer 635. It also decodes the quantized value and outputs it to the pitch recovery filter 850.

The pitch recovery filter 850 receives the sound source signal v(n
) and output it.

V(n)=c(n)+b'・v(n-M)
(7) where c(n) is the selected noise signal.

The synthesis filter 860 manually calculates v(n) to obtain and output synthesized speech.

Switch 865 outputs the output of adder 627 to subtracter 535 in voiced frames, and outputs the output of synthesis filter 860 in unvoiced frames.

On the receiving side, a decoding circuit 875 decodes the gain and index of the noise signal.

Parameter decoding circuit 870 decodes pitch gain b', pitch period M', and spectral parameters a and l.

The pitch recovery filter 880 performs the same operation as the pitch recovery filter 850 on the transmission side, and decodes the sound source signal in the unvoiced frame.

A switch 870 switches the sound source signal between voiced frames and unvoiced frames.

This concludes the explanation of one example of the second invention.

The configuration described above is only one embodiment of the present invention, and various modifications are possible.

As a method for calculating multi-pulses, in addition to the method shown in Document 1, various well-known methods can be used.

For example, “A 5tu” by Ozawa et al.
dy on Pulse 5earch Algori
thms for Multi-pulse 5pee
ch Coder Realization” (IE
EE JSAC, pp.

133-141.1986X Document 7).

In addition, as a method for calculating the pitch period and pitch gain, in addition to the method shown in the above embodiment, for example, as shown in the following equation (8), the past sound source signal v(n) and the pitch reproduction filter, It is also possible to search for a position M that minimizes the error power E between the signal reproduced by the filter and the input audio signal x(n) of the current subframe, and find the coefficients at that time.

E=Σ[(x(n)-b-V(n-T)*h(n)) w(n)] (8) Here, h(n) is the impulse response of the synthesis filter, w(n ) shows the impulse response of the auditory weighting circuit.

Further, if the transmitting side synthesis filter 610 is configured to reproduce the weighted signal and the weighted signal is subtracted from the weighted signal, the weighted signal can be omitted.

Also, the synthesis filter 610.625.86 on the transmitting side
0 can also be made common.

Furthermore, the subtraction of the influence signal can be omitted on the transmitting side, although the characteristics are slightly degraded. With such a configuration, the subtracter 535, the synthesis filter 625, the adder 627, the pitch recovery filter 850, and the synthesis filter 860 become unnecessary, and the configuration can be simplified.

(Effects of the Invention) According to the first invention, in voiced frames, pulses with strong periodicity for each pitch can be expressed very efficiently by determining pulses in one subframe section by pitch interpolation, and Since multi-pulses are obtained without using pitch interpolation for pulses whose correlation is not very strong, compared to the conventional method that uses pitch prediction for all pulses, it is possible to obtain multi-pulses without using pitch interpolation. This has the effect of greatly improving the sound quality in areas where the periodicity is slightly weakened. Furthermore, since pitch interpolation requires multipulses only for one subframe,
This has the great effect of significantly reducing the amount of calculation required compared to pitch prediction multi-pulse. Furthermore, according to the second invention, in unvoiced frames where there is no periodicity and the sound source signal is noise, the best noise signal is selected to represent the sound source, so the sound quality is further improved compared to the conventional method. effective.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the audio encoding/decoding method according to the first invention, and FIG. 2 is a block diagram showing the configuration of an embodiment of the audio encoding/decoding method according to the second invention. 3 are block diagrams showing the operation of the present invention. FIG. 4 is a block diagram showing an example of pitch interpolation multi-pulse. FIG. 5 is a block diagram showing an example of a conventional method. In the figure, 150...LPG, pitch analysis section, 25
0... Sound source pulse calculation unit, 270... Pulse meter 8 unit, 520, 522... Spectrum, pitch parameter calculation circuit, 525... Parameter quantizer, 530...
・Inverse quantizer, 535.260...Subtractor, 540・
...Weighting circuit, 550.0. Impulse response calculation circuit, 560...Autocorrelation function calculation circuit, 570.60
3... Cross-correlation function calculation circuit, 585.620...
Quantizer, 627...Adder, 586...Pulse calculation circuit, 605.726...Pitch interpolation circuit, 610
．． 625.760.860...Synthesis filter, 635
... multiplexer, 710 ... demultiplexer, 720 ... first pulse decoder, 725 ... second
pulse decoder, 750.870...parameter decoder, 727...pulse generator, 800...codebook, 810...convolution circuit, 820.825.8
65... Switch, 830... Signal selection circuit, 85
0.880...Pitch reproduction filter, 875...Decoding circuit.

Claims (2)

[Claims] (1) On the transmitting side, a discrete audio signal is input, a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch period are extracted from the audio signal for each frame, and the audio signal of the frame is adjusted according to the pitch parameter. dividing the audio signal into small sections, using the pitch parameter and the spectrum parameter to obtain a first multi-pulse for the audio signal in one of the small sections, and coefficients for correcting the multi-pulse in other sections. After removing the signal obtained from the multi-pulse and the coefficient from the audio signal, a second multi-pulse is obtained using the spectral parameter, and on the receiving side, the first multi-pulse, the pitch parameter, and the signal are removed from the audio signal. A speech encoding/decoding method characterized in that a sound source signal is restored using a correction coefficient and the second multi-pulse, and a synthesized speech signal is obtained by driving a synthesis filter configured using the spectral parameter. . (2) On the transmitting side, a discrete audio signal is input, a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch period are extracted from the audio signal for each frame, and the audio signal of the frame is adjusted according to the pitch parameter. dividing the audio signal into small sections, and calculating a first multipulse using the pitch parameter and the spectrum parameter in one of the small sections as a sound source signal of the audio signal, and correcting the multipulse in other sections. a multipulse sound source obtained by calculating a second multipulse using the spectral parameter for the signal obtained by removing the multipulse and the signal calculated by the coefficient from the audio signal, It is expressed using a noise signal selected from a codebook consisting of predetermined types of noise signals so as to reduce the error power between the speech signal and a composite signal obtained from the noise signals, and the reception side 1 multipulse, the pitch parameter, the correction coefficient, and the second multipulse, or restore the sound source signal using the selected noise signal and configure using the spectral parameter. A speech encoding/decoding method characterized in that a synthesized filter is driven by the sound source signal to obtain a synthesized speech signal.