JP2853170B2 - Audio encoding / decoding system - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an audio encoding / decoding system for efficiently encoding and decoding an audio signal at a low bit rate.

(Prior Art) As a method of transmitting an audio signal at a low bit rate, for example, about 16 Kb / s or less, a multi-pulse encoding method and the like are known. In these, the sound source signal is represented by a plurality of pulse combinations (multi-pulse), the characteristics of the vocal tract are represented by a digital filter, and the information of the sound source pulse and the coefficient of the filter are obtained and transmitted for each fixed time section (frame). I have. For details of this method, see, for example, Araseki, Ozawa, Ono, Ochi
“Multi-pulse Excited Speech Coder Base by ai
d on Maximum Crosscorrelation Search Algorithm ",
(GLOBECOM83, IEEE Global Telecommunication, lecture number 23.3, 1983) (Reference 1). In this method, a good voice signal can be output after decoding by separating and expressing the vocal tract information and the sound source signal and using a combination (multi-pulse) of a plurality of pulse trains as means for expressing the sound source signal. . The basic concept of obtaining a pulse train representing a sound source signal will be described with reference to FIG. An audio signal divided for each frame is input from an input terminal 900 in the figure. The spectrum parameters obtained from the audio signal of the current frame are input to the synthesis filter 920. An initial multi-pulse is generated in the sound source calculation circuit 910 and is input to the synthesis filter 920 to obtain a synthesized speech waveform as an output. A subtractor 940 subtracts a synthesized speech waveform from the input signal. The result is input to the weighting circuit 950 to obtain the weighted error power in the current frame. Then, the sound source calculation circuit 910 obtains the amplitudes and positions of a specified number of multi-pulses so as to minimize the weighted error power.

(Problems to be Solved by the Invention) However, in this conventional method, the sound quality is good when the bit rate is sufficiently high and the number of sound source pulses is sufficient, but the sound quality deteriorates as the bit rate is reduced. The problem fits.

In order to improve this problem, a pitch prediction multipulse method using quasi-periodicity (pitch correlation) for each pitch of a multipulse sound source has been proposed. For more information on this method,
For example, the detailed description is omitted from the specification of Japanese Patent Application No. 58-139022 (Reference 2). However, the quasi-periodicity of each pitch of a multi-pulse sound source is considered to be large for large-amplitude pulses, but not all pulses have such periodicity. Is considered to be small. In the pitch prediction multi-pulse method described in Reference 2, all pulses are obtained by pitch prediction assuming periodicity for each pitch for all predetermined numbers of pulses in a frame. There was a problem that the characteristics of the pulse deteriorated due to the pitch prediction.
In particular, this is remarkable in a transition section between vowels and a transition section, and there is a problem that sound quality is deteriorated in such a section.

Further, in the method of Document 2, since the pitch information is included in the impulse response, an impulse response having a very long time length (for example, 20 msec or more) is required, and a predetermined number of all the pulses are determined by pitch prediction. Therefore, the amount of calculation required for searching for a pulse is extremely large, and it is difficult to realize a compact device even with current LSI techniques.

It is an object of the present invention to provide a speech encoding / decoding method which can reproduce a better sound than before even when the bit rate is high or lower, and which can be realized with a small amount of calculation. is there.

(Means for Solving the Problems) According to a speech encoding / decoding method of the present invention, a discrete speech signal is input on the transmission side, and a spectrum parameter representing a spectrum envelope and a pitch representing a pitch period are provided for each frame from the speech signal. Parameters, and the audio signal of the frame is divided into small sections according to the pitch parameter,
A first multi-pulse is obtained using the pitch parameter and the spectrum parameter for an audio signal of one of the small sections, and a coefficient for correcting the multi-pulse is obtained in another section, and the multi-pulse is obtained. And a signal obtained by removing the signal obtained from the coefficient from the audio signal is used to obtain a second multi-pulse using a spectrum parameter. On the receiving side, the first multi-pulse, the pitch parameter and the coefficient are obtained. And recovering the sound source signal using the second multi-pulse, and further driving a synthesis filter configured using the spectrum parameters to obtain a synthesized speech signal.

Also, in the speech encoding method according to the present invention, a discrete speech signal is input on the transmission side, and a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch period are extracted from the speech signal for each frame, and the speech of the frame is extracted. The signal is divided into small sections according to the pitch parameter, and one of the small sections is used as a sound source signal of the audio signal.
In one section, a first multi-pulse is obtained using the pitch parameter and the spectrum parameter, and in another section, a coefficient for correcting the multi-pulse is obtained, and a signal obtained from the multi-pulse and the coefficient is obtained from the audio signal. A multi-pulse sound source obtained by obtaining a second multi-pulse using the spectrum parameter for a signal obtained by removing the signal, or synthesizing the speech signal from a codebook composed of a predetermined type of noise signal. The signal is represented by using a noise signal selected so as to reduce the error power with respect to the signal, and on the receiving side, the source signal is restored using the first multi-pulse, the pitch parameter, the coefficient, and the second multi-pulse. Alternatively, a sound source signal is reconstructed using the selected noise signal, and a synthesis file configured using the spectrum parameters is restored. The filter is driven by the excitation signal and obtaining the synthesized speech signal.

(Operation) In the speech encoding / decoding method according to the first aspect of the invention, a sound source signal of a speech signal in a frame section (for example, 20 ms) is converted into a multipulse (Pulse interpolation) obtained by pitch interpolation in a small section obtained by dividing a frame in a sound section. It is characterized by using a multi-pulse (second multi-pulse) obtained without pitch prediction in the entire frame (first multi-pulse). The calculation of the first multipulse is performed as follows. In order to use the quasi-periodicity of each pitch of the multi-pulse sound source very efficiently and greatly reduce the amount of calculation, the frame is divided into small sections (sub-frames) corresponding to the pitch cycle in advance, and A multipulse is obtained only for one subframe (representative section). For other sub-frames, a correction coefficient for correcting the gain and phase of the multi-pulse obtained in the representative section is obtained, and in this other sub-frame, the gain and phase of the multi-pulse in the representative section are corrected using this coefficient. To generate a pulse to restore the pulse of the entire frame. Then, after the signal is reproduced in a frame by the pulse and the signal is subtracted from the audio signal, a multi-pulse (second multi-pulse) is obtained in the frame by the same method as in Reference 1.

The basic processing of this method will be described below with reference to FIG. FIG. 3 is a block diagram showing the operation of the present invention.
An audio signal is input from the input terminal 100, and the audio signal is divided into frames of a predetermined time length (for example, 20 ms). The LPC / pitch analysis unit 150 obtains LPC coefficients of a predetermined order as spectral parameters representing a spectral envelope from the audio signal of the frame by a well-known LPC analysis. As the LPC coefficient, in addition to the linear prediction coefficient a _i used here, other good parameters such as LSP, formant, and LPC cepstrum can also be used. In addition, analysis methods other than LPC, for example, cepstrum, PSE, ARMA method and the like can also be used. Hereinafter, description will be made assuming that a linear prediction coefficient is used. Also, 150 calculates a pitch period M as a pitch parameter from the voice of the frame. A well-known autocorrelation method can be used for this.

The operation of the pitch interpolation multi-pulse calculation unit 250 and the multi-pulse calculation unit 270 will be described with reference to FIG. FIG. 4A shows an audio signal of a frame. Here, the frame length is set to 20 ms as an example. The pitch interpolation multi-pulse calculation unit 250 first divides a frame into small sections (subframes) using a pitch period M as shown in (b). Here, the length of the subframe is the same as the pitch period M.

Next, the impulse response h of the synthesis filter composed of the linear prediction coefficients is calculated in the same manner as in Reference 1.
(N) auto-correlation function R _hh (m), cross-correlation function Φ _hx between the perceptually weighted audio signal and the impulse response h (n)
(M) is obtained. Next, only for a predetermined section (hereinafter referred to as a representative section, here, for example, the section in FIG. 4B) of the subframe, a predetermined number K (here, 4 amplitude g _i of the multi-pulse (first multi-pulse) of are) obtains the position m _i.
Here, the method of obtaining the multi-pulse can be referred to the above-mentioned document 1. FIG. 4 (c) shows the obtained multi-pulse. next,
In subframes other than the representative section, a gain correction coefficient and a phase correction coefficient for generating a pulse by correcting the gain and phase of the multipulse obtained in the representative section are obtained. Gain correction coefficient in the j-th subframe in the frame
c _j and the phase correction coefficient d _j are determined so as to minimize the error power of the following equation.

Here, x _j (n) and s _j (n) indicate the speech signal in the j-th subframe and the synthesized speech obtained by correcting the gain and phase of the multi-pulse, respectively. However Here, h (n) is the impulse response of the synthesis filter. By placing a 0 to partial differentiation (2) by substituting expression of equation (1) c _j, it can be calculated c _j, d _j that minimizes equation (1). For details, reference can be made to Japanese Patent Application No. 63-208201 (Document 3). Thus, basically, the gain correction coefficient and the phase correction coefficient are obtained for all the other subframe sections in the frame. Then, the pulses of the entire frame are reproduced as shown in FIG. 4D using the multi-pulse of the representative section, the gain correction number, and the phase correction coefficient. The position of the representative section in the frame may be determined by searching for some subframes or may be determined in advance. The details of the former method can be referred to, for example, the above-mentioned Document 3.

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

Here, a _i is a linear prediction coefficient.

The subtractor 260 subtracts x '(n) from the audio signal x (n) according to the following equation to obtain e (n).

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

On the other hand, in an unvoiced frame, the amplitude and position of the multipulse are obtained for the entire frame.

The transmission information on the transmitting side includes, in addition to the spectral parameters of the synthesis filter, in a voiced frame, the spectral parameters a _i representing the spectral envelope, the pitch M, the amplitude and position of the K multipulses in the representative section, the gain correction coefficient, the phase The correction coefficient, the position in the frame of the representative section, and the amplitude and position of the Q multi-pulses. In an unvoiced frame, the multipulse amplitude and position are transmitted.

In the second invention, in a voiced frame, the same operation as in the first invention is performed, but in an unvoiced frame, not a multipulse but a noise signal in which one type is selected from a codebook including a predetermined type of noise signal is used. The sound source signal is represented by using the sound source signal. As the noise signal, for example, a random number having a Gaussian statistical distribution can be used. The length (the number of dimensions) of the noise signal in the time direction is shorter than the normal frame (for example, 5 to 10 ms). The type of noise signal is 2 ^B
Type. As a method of selecting the best noise signal for the input speech from such a codebook, a noise filter is used to drive a synthesis filter to synthesize speech, obtain an error power from the original speech, and calculate the error power. A method for selecting a noise signal to be minimized is known. For details of this method, for example, Sc
“Code-excited linear predict by hroeder, Atal
ion (CELP): High quality speech at very low bitrat
es "(Proc. ICASSP, pp. 937-940, 1985)
(Reference 4) can be referred to.

In the unvoiced frame, the index indicating the selected noise signal, the gain, the pitch gain of the pitch reproduction filter,
The pitch period and the spectrum parameters of the 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 spectrum and pitch parameter calculation circuit 520 calculates the spectrum parameters a _i of the synthesis filter representing the audio signal spectrum envelope of the divided frame section (for example, 20 ms),
Determined by the well-known LPC analysis method. The pitch period M
Is calculated by the well-known autocorrelation method.

The quantizer 525 performs quantization on the obtained spectrum parameter and pitch period. As a quantization method, scalar quantization or vector quantization as described in Japanese Patent Application No. 59-272435 (Reference 5) may be performed. For a specific method of vector quantization, see, for example, “Vector quantizati
on in speech coding "(Proc.IEEE.pp.1551-1558,198
5) Refer to papers such as (Reference 6).

The inverse quantizer 530 inversely quantizes using the result of the quantization and outputs the result.

The subtractor 535 subtracts the influence signal from the audio signal of the frame and outputs the result.

The weighting circuit 540 performs perceptual weighting on the audio signal using the inversely quantized spectral parameter. The weighting method is described in the weighting circuit 200
Can be referred to.

The impulse response calculation circuit 550 calculates the impulse response h (n) of the synthesis filter that has been subjected to the perceptual weighting using the dequantized spectral parameters a ′ _i . The specific method can be referred to the impulse response calculation circuit of the above reference 2.

The autocorrelation function calculation circuit 560 calculates an autocorrelation function R _hh (m) for the impulse response, and outputs the results to the sound source pulse calculation circuit 580 and the pulse calculation circuit 586, respectively. The calculation method of the autocorrelation function is described in the document 2
180 can be referenced.

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

The sound source pulse calculation circuit 580 first divides a frame into subframe sections as shown in FIG. 4B using a pitch period M 'obtained by dequantizing a frame. Then, for one predetermined subframe section (representative section) (for example, the subframe in FIG. 4B), Φ _xh (m) and R _hh
(M) and using the determined amplitude g _i and position m _i of the K multi-pulse train (first multi-pulse). For the method of calculating the pulse train, reference can be made to the sound source pulse calculation circuit of Reference 2.

In the correction coefficient calculation circuit 583, it is shown in the operation section (1),
According to the equation (2), the gain correction coefficient c _j and the phase correction coefficient _dj are calculated and output in the subframe sections other than the representative section.

The quantizer 585 quantizes the amplitude and position of the multi-pulse train and outputs a code. The specific method can be referred to the above-mentioned documents 1 and 2. In addition, a code is output by quantizing the gain correction coefficient, the phase correction coefficient, and the position in the frame of the representative section. The specific method can be referred to, for example, the above-mentioned document 3. These outputs are further dequantized and output to the 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 'according to the above equation (3).
(N) is obtained.

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

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

The 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 obtains the amplitude and position of a predetermined number of multi-pulses (second multi-pulses) using the cross-correlation function and the auto-correlation function of the impulse response h (n).

The quantizer 620 quantizes and outputs the amplitude and position of the multi-pulse, and inversely quantizes these and outputs the result to the synthesis filter 625.

 The combining filter 625 combines and outputs the residual signal.

An adder 627 adds the outputs of the synthesis filters 625 and 610 to obtain a reproduced signal of the frame, and further obtains and outputs an influence signal for the next frame. Reference 2 can be referred to for a specific method of calculating the influence signal.

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

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

The first pulse decoder 720 decodes the amplitude and position of the pitch interpolation multi-pulse. The second pulse decoder 725 is
The multi-pulse amplitude and position are decoded. The parameter decoder 750 works in the same way as the inverse quantizer 530 on the transmission side,
The spectrum parameter a ′ _i and the pitch period M ′ are decoded and output.

The pitch interpolation circuit 726 performs the same operation as the pitch interpolation circuit 605 on the transmission side.

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

The adder 740 adds the output signals of the pulse generator 727 and the pitch interpolation circuit 726 to obtain a driving sound source signal of the frame, and drives the synthesis filter circuit 760.

The synthesis filter circuit 760 obtains and outputs a synthesized speech waveform for each frame using the driving excitation signal and the decoded spectrum parameter.

 This is the end of the description of the embodiment of the first invention.

FIG. 2 is a block diagram showing one embodiment of the second invention. In the figure, components having the same reference numerals as those in FIG. 1 perform the same operations as those in FIG.

In the figure, spectrum and pitch parameter calculation circuit
In step 522, a spectrum parameter a _i is obtained by using a well-known LPC analysis, and a pitch period M and a pitch gain b are obtained as pitch parameters by using a well-known autocorrelation method.

The quantizer 522 converts the spectral parameters a _i into PARCOR coefficients or LSP coefficients and then quantizes them. Here, the PARCOR coefficient is used. Further, the pitch period M and the pitch gain b are quantized. Further, these quantized values are decoded to output decoded values a ′ _i , M ′, b ′.

The codebook 800 stores 2 ^B (B indicates the number of bits) kinds of noise signals in advance. For the method of generating a noise signal, reference can be made to the aforementioned reference 4. From these, the signals are output to the convolution circuit 810 one by one.

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 for voiced frames, and outputs the output of the convolution circuit 810 to the autocorrelation function calculation circuit 560 for unvoiced frames. Here, voiced or unvoiced can be determined as voiced when the value of the decoded pitch gain b 'exceeds a predetermined threshold, and unvoiced otherwise.

The switch 825 outputs the output of the autocorrelation function calculation circuit 560 to the sound source pulse calculation circuit 580 for voiced frames, and outputs to the signal selection circuit 830 for unvoiced frames.

The signal selection circuit 830 has a cross-correlation function Φ _xh and an auto-correlation function R
_The following equation is calculated using _hh .

G = (Φ _xh ) ² / R _hh (6) The equation (6) is calculated for all noise signals, and (6)
A noise signal that maximizes the expression is selected, and an index representing the selected noise signal and the gain G obtained by Expression (6) are output.

The encoder 840 quantizes the gain G with a predetermined number of bits and outputs the result to the multiplexer 635. Further, it decodes the quantized value and outputs it to pitch reproduction filter 850.

The pitch reproduction filter 850 generates the sound source signal v according to the following equation.
(N) is obtained and output.

V (n) = c (n) + b ′ · v (n−M) (7) where c (n) is the selected noise signal.

The synthesis filter 860 receives v (n), obtains and outputs synthesized speech.

The switch 865 outputs the output of the adder 627 for the voiced frame to the subtractor 535, and outputs the output of the synthesis filter 860 for the unvoiced frame.

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

The parameter decoding circuit 870 decodes the pitch gain b ', the pitch period M', and the spectrum parameter _ai '.

Pitch reproduction filter 880 performs the same operation as pitch reproduction filter 850 on the transmission side, and decodes a sound source signal in an unvoiced frame.

The switch 870 switches a sound source signal between a voiced frame and an unvoiced frame.

 This is the end of the description of the embodiment of the second invention.

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

As a method of calculating the multi-pulse, various well-known methods can be used in addition to the method shown in the above-mentioned document 1. This includes, for example, “A Study on
Pulse Search Algorithms for Multi-pulse Speech Co
der Realization "(IEEE JSAC, pp. 133-141, 1986) (Reference 7).

Also, pitch period and pitch gain can be calculated by:
In addition to the method shown in the above embodiment, for example, the following (8)
As shown in the equation, a position M that minimizes the error power E between the past sound source signal v (n), the signal reproduced by the pitch reproduction filter and the synthesis filter, and the input audio signal x (n) of the current subframe. And the coefficient b at that time can be obtained.

Where h (n) is the impulse response of the synthesis filter,
w (n) indicates the impulse response of the auditory weighting circuit.

In addition, if the weighting signal is reproduced by the synthesis filter 610 on the transmission side and is subtracted from the weighting circuit 540, the weighting circuit 600 can be omitted.

Also, the synthesis filters 610, 625, and 860 on the transmission side can be shared.

In addition, although the characteristics are slightly lowered, the subtraction of the influence signal on the transmitting side can be omitted. With such a configuration,
The subtractor 535, the synthesis filter 625, the adder 627, the pitch reproduction 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 a voiced frame, a pulse having a strong periodicity for each pitch is represented very efficiently by obtaining a pulse in one subframe section by pitch interpolation, Since the multi-pulse is calculated without using the pitch interpolation for the pulse whose correlation is not so strong for each pulse, the vowel transition part and the transition part are compared with the conventional method that uses the pitch prediction for all the pulses. There is an effect that the sound quality can be greatly improved in a portion where the periodicity is slightly weakened. Further, in the pitch interpolation, a multi-pulse is obtained only for one sub-frame, so that there is a great effect that a necessary calculation amount can be greatly reduced as compared with the pitch prediction multi-pulse. Further, according to the second aspect, in an unvoiced frame in which the sound source signal is noisy without periodicity, the best noise signal is selected to represent the sound source, so that the sound quality is further improved as compared with the conventional method. effective.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a speech encoding / decoding system according to the first invention, and FIG. 2 is a block diagram showing a configuration of an embodiment of a speech encoding / decoding system according to the second invention. FIG. 3 is a block diagram showing the operation of the present invention.
FIG. 4 is a block diagram showing an example of a pitch interpolation multi-pulse. FIG. 5 is a block diagram showing an example of the conventional system. In the figure, 150 LPC, pitch analyzer, 250 pulse generator, 270 pulse calculator, 520, 522 spectrum,
Pitch parameter calculation circuit, 525: parameter quantizer, 530: inverse quantizer, 535,260: subtractor, 540: weighting circuit, 550: impulse response calculation circuit, 560: autocorrelation function calculation circuit, 570, 603: cross correlation Function calculation circuit, 585,6
20 ... Quantizer, 627 ... Adder, 586 ... Pulse calculation circuit, 60
5,726: pitch interpolation circuit, 610, 625, 760, 860: synthesis filter, 635: multiplexer, 710: demultiplexer, 72
0 ... first pulse decoder, 725 ... second pulse decoder, 75
0,870… Parameter decoder, 727… Pulse generator, 800…
Codebook, 810: convolution circuit, 820, 825, 865 switch, 830: signal selection circuit, 850, 880: pitch reproduction filter, 875: decoding circuit.

──────────────────────────────────────────────────続 き Continued on the front page (58) Fields investigated (Int. Cl. ⁶ , DB name) G10L 3/00-9/18 H03M 7/30 H04B 14/04 JICST (JOIS)

Claims (2)

(57) [Claims] 1. A transmitting side inputs a discrete voice signal, extracts a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch period for each frame from the voice signal, and converts the voice signal of the frame into the pitch parameter. , A first multi-pulse is obtained for the audio signal of one of the small sections using the pitch parameter and the spectrum parameter, and the multi-pulse is corrected in another section. The second multi-pulse is obtained by using the spectrum parameter after removing the signal obtained by the multi-pulse and the coefficient from the audio signal, and the receiving side obtains the first multi-pulse and the pitch parameter. And recovering the sound source signal using the correction coefficient and the second multi-pulse. Speech coding and decoding method and obtains the synthesized speech signal by driving a configured synthesis filter using the Le parameters. 2. A transmitting side inputs a discrete voice signal, extracts a spectrum parameter representing a spectrum envelope and a pitch parameter representing a pitch period for each frame from the voice signal, and converts the voice signal of the frame into the pitch parameter. , And a first multipulse is obtained as a sound source signal of the audio signal using the pitch parameter and the spectrum parameter in one of the small sections, and the multipulse is obtained in another section. A multi-pulse sound source obtained by obtaining a second multi-pulse using the spectral parameter with respect to a signal obtained by removing the multi-pulse and a signal obtained by the coefficient from the audio signal. Alternatively, the speech signal and the noise may be obtained from a codebook composed of a predetermined type of noise signal. It is expressed by using a noise signal selected so as to reduce the error power with respect to the synthesized signal obtained from the signal, and the receiving side uses the first multi-pulse, the pitch parameter, the correction coefficient, and the second multi-pulse. Or restore the sound source signal using the selected noise signal, and restore the sound source signal using the selected noise signal, and drive a synthesis filter configured using the spectral parameters with the sound source signal to obtain a synthesized speech signal. Audio encoding / decoding system to be used.