KR19990061574A - Multi-pulse excitation linear prediction encoding / decoding method and apparatus therefor - Google Patents

Abstract

The present invention provides a method and apparatus for multi-pulse excitation linear prediction encoding / decoding that can reduce a generated bit rate by efficiently encoding a prediction error.

To this end, a coding method according to the present invention is a coding method for coding a speech signal, comprising: A LPC analysis filtering step of receiving a speech signal and obtaining and outputting a LPC code and a prediction error signal; And a coding step of calculating and outputting the position of the peak value and the magnitude of the peak value with respect to the prediction error signal from the LPC analysis filter. The decoding method according to the present invention is a decoding method in which a coding apparatus encodes an original speech signal using an LPC code and a distance information between neighboring peak values of a prediction error signal and size information of a peak value, In a decoding method for decoding into a speech signal; Calculating a difference signal and a step size of neighboring peak values based on size information of the peak value and distance information, and decoding the prediction error signal using the difference signal and the step size; And a synthesis filtering step of synthesizing the predictive error signal of the decoding step with the linear predictive coding coefficient and outputting the original speech signal.

Description

Multi-pulse excitation linear prediction encoding / decoding method and apparatus therefor

The present invention relates to a multi-pulse excitation linear prediction (MPELP) encoding / decoding method and apparatus, and more particularly, to a multi-pulse excitation linear prediction encoding / decoding method and apparatus for a speech signal. will be.

In the multimedia age, speech signal is one of important information sources with image and text information, and it is relatively difficult to obtain a high compression rate because redundancy is relatively small compared with other image signals and audio signals. Yo

In order to compress such a speech signal, three methods are generally used. One is a waceform coding method in which only the signal itself is processed in units of samples on a time domain without considering a generation model of the source signal, There are a parametric coding scheme for modeling the characteristics of a sound source and extracting only specific parameters, and finally a hybrid coding scheme in which a waveform coding scheme and a parametric coding scheme are mixed.

Representative waveform encoding schemes include a differential pulse code modulation (DPCM) scheme, an adaptive differential pulse code modulation (ADPCM) scheme, and an adaptive delta modulation (ADM) scheme . Such a waveform coding method is relatively natural in sound quality, but requires a high bit rate of 32K to 22Kbps (bit per second).

The difference pulse code modulation (DPCM) scheme uses a fact that the average level change of adjacent samples is relatively small, and quantizes the difference between the immediately previous sample value and the current sample value into a multi-bit code (PCM) method. However, there is a disadvantage in that sufficient response characteristics can not be obtained when the signal level is abruptly changed.

The adaptive differential pulse code modulation (ADPCM) scheme uses a quantization step width of amplitude variation to compensate for the disadvantages of the differential code modulation (DPCM) scheme. That is, by adaptively quantizing the difference signal between the predicted value and the input signal without directly quantizing the input signal, the bit rate can be saved with an intermediate sound quality.

The adaptive delta modulation (ADM) scheme is a method that adaptively changes the step size, and adds or subtracts the step size as needed to approach the original signal waveform. However, this method can not keep up with the sudden change of the original signal, so that slope overload distortion occurs, and even in a portion where the original signal change is slow or the change is very small, a granular noise There is a problem that reproduction to the original signal waveform becomes difficult.

The parametric coding method is a linear predictive coding (LPC) method. The linear predictive coding method uses signal correlation in the time axis direction. A speech signal has a relatively small change in amplitude (sound pressure level) with respect to time, and therefore there is a strong correlation between successive signals. Therefore, in the predictive encoding, the data rate is compressed by predicting the current sample in the sample that existed in the past and encoding the prediction error which is the difference between the actual data and the predictive data. Herein, a method of LPC (LPC) coding for predicting a current sample while continuously updating the previous 10 samples and LPC 20 for performing LPC prediction with the previous 20 samples is referred to as LPC20. However, the linear predictive coding has a disadvantage in that the compression rate of data is high but the sound quality is not relatively natural.

In order to overcome these drawbacks, a hybrid coding scheme is used. In this scheme, a residual excitation linear prediction (RELP) scheme is used.

As shown in FIG. 1, the RELP encoding apparatus employing the RELP scheme receives an input X (z) of a speech signal through the frame buffer 10 and outputs a LPC coefficient Ai and a predictive error signal signal; E (z)] of the prediction error signal E (z), a prediction error signal E (z) having a small dynamic range, ) Outputted from the waveform encoder 14. The waveform encoder 14 outputs a waveform signal.

The LPC analysis filter 12 includes an LPC analyzer 12A for analyzing a speech signal input [X (z)] input through the frame buffer 10 by a linear prediction method and calculating an LPC coefficient Ai, And an inverse filter 12B for outputting a prediction error E (z) by filtering the speech signal input X (z) using the coefficient Ai as a filter coefficient. Here, the prediction error [E (z)] has a relatively flat spectrum, but most important frequency components for improving the speech quality are included in the frequency range of 1000 Hz or less.

The waveform encoder 14 is for reducing the bit rate by eliminating relatively less important high frequency components and includes a low pass filter 14A for low-pass filtering the prediction error E (z) with a band width of approximately 1000 Hz A decimator 14B for decimating the filtering output, a DFT unit 14C for converting the decimated signal into a frequency domain through a discrete Fourier transform (DFT) .

The LPC coefficient Ai of the LPC analysis filter 12 and the prediction error C (E (z)) encoded by the waveform encoder 14 are channel-coded by a channel encoder not shown and output as a transmission channel.

2, the RELP decoding apparatus decodes the LPC coefficient Ai and the encoding prediction error C (E (z)) channel-decoded in the channel decoder (not shown) And decodes the decoded prediction error E (z) by receiving a coding prediction error C (E (z)) through the frame buffer 20, And a predictive error E (z) of the waveform decoder 22 are synthesized by combining the LPC coefficient Ai through the frame buffer 20 and the predictive error E (z) And an LPC synthesis filter 24 for outputting the speech signal [X (z)].

Here, the waveform decoder 22 converts the encoding prediction error [C (E (z))] from the frequency domain to the time domain by an inverse discrete Fourier transform (IDFT). Since the converted signal does not include a high frequency component, a high-frequency component is generated by a full-wave rectifier 22B, and only a high-frequency component is extracted through filtering by a high-pass filter 22C.

Subsequently, the adder 22D adds the low-frequency component of the encoding prediction error [C (E (z)]] output from the IDFT unit 22A and the high-frequency component that has passed through the high-pass filter 22C, E (z)].

The LPC synthesis filter 24 synthesizes the original speech signal X (z) by combining the LPC coefficient Ai through the frame buffer 20 and the prediction error E (z) of the waveform decoder 22, ].

As described above, in the conventional RELP encoding apparatus, the LPC coefficient and the prediction error are obtained by passing the input of the speech signal through the LPC analysis filter. In addition, since the dynamic range of the prediction error signal is small, a prediction error is obtained by compressing the prediction error using a waveform encoder, and the LPC coefficient and the encoded prediction error are transmitted. The conventional RELP decoding apparatus receives the LPC coefficient and the encoded prediction error and restores the original speech signal X (z).

However, the above-mentioned conventional RELP unit treble decoding apparatus has a problem that it has a high bit rate of about 16K bps as a whole because it uses an existing waveform coding technique with a high bit rate. That is, since data of all samples is encoded and transmitted, there is a problem that the amount of information to be transmitted is large.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and it is an object of the present invention to provide a multi-pulse excitation linear prediction encoding / decoding method capable of efficiently reducing a generated bit rate by efficiently encoding a prediction error.

Another object of the present invention is to provide a multi-pulse excitation linear prediction protection / decoding apparatus capable of efficiently reducing a generated bit rate by efficiently encoding a prediction error.

Another object of the present invention is to provide a novel hybrid protection / decoding method using a Peak Value and Distance Algorithm (PVDA), and to provide a multi-pulse excitation linear And to provide a predictive encoding / decoding method and apparatus therefor.

According to an aspect of the present invention, there is provided a method for encoding a speech signal, the method comprising: A LPC analysis filtering step of receiving a speech signal and obtaining and outputting a LPC code and a prediction error signal; And a coding step of calculating the position of the peak value and the magnitude of the peak value with respect to the prediction error signal from the LPC analysis filtering step and outputting the result.

The encoding step includes a peak value calculation step of receiving the prediction error signal and calculating a peak value of the signal and outputting the size information and the position information of the peak value; And calculating a distance between adjacent peaks based on the position information and outputting the distance.

In order to attain the above object, the present invention provides a multi-pulse excitation linear predictive decoding method for encoding an audio signal in a coding apparatus, the linear predictive coding method comprising the steps of: A decoding method of decoding information of an original speech signal by receiving size information of information and a peak value, the decoding method comprising: And a synthesis filtering step of synthesizing the linear prediction code coefficient and the prediction error signal of the decoding step and outputting the original speech signal.

The decoding step includes a difference signal calculating step of calculating a difference signal of adjacent peak values based on the size information; A step size calculation step of calculating a step size using the distance information and a difference signal from the difference signal calculation step; A signal waveform reproducing step of reproducing the signal waveform by linear interpolation using the difference signalizing step size; A conversion step of converting a reproduction waveform of the reproduction step of the signal waveform reproduction step into an analog signal; And a filtering step of performing low-pass filtering on the analog waveform signal from the conversion step.

According to another aspect of the present invention, there is provided a multi-pulse excitation linear prediction encoding apparatus for encoding a speech signal, the apparatus comprising: A LPC analysis filter means for receiving a speech signal and obtaining and outputting a LPC coefficient and a prediction error signal; And a coding means for obtaining the position of the peak value and the magnitude of the peak value with respect to the prediction error signal from the LPC analysis filter means and outputting the magnitude.

Here, the encoding means includes: peak value calculation means for receiving the error signal, computing a peak value of the signal, calculating size information of the peak value, and outputting the position information; And distance calculating means for calculating and outputting a distance between adjacent peaks based on the position information.

According to another aspect of the present invention, there is provided a multi-pulse excitation linear prediction decoding apparatus for decoding an audio signal in a coding apparatus, the apparatus comprising: And decoding information of the distance information and size information of the peak value to the original voice signal, the apparatus comprising: Decoding means for calculating a difference signal and a step size of adjacent peak values based on size information of the peak value and distance information and decoding the prediction error signal using the difference signal and the step size; And synthesis filter means for synthesizing the linear prediction code coefficient and the prediction error signal of the decoding means to output an original speech signal.

Here, the decoding means may comprise: a difference signal calculation means for calculating a difference signal of adjacent peak values based on the size information; A step size calculation means for calculating a step size using the distance information and a difference signal from the difference signal calculation means; A signal waveform reproducing means for reproducing a signal waveform by linear interpolation using the difference signal and the step size; Conversion means for converting the reproduction waveform of the signal waveform recoding means into an analog signal; And filtering means for performing low-pass filtering on the analog waveform signal from the conversion means.

FIG. 1 is a block diagram schematically showing a conventional redundant excitation linear prediction (RELP) encoding apparatus,

FIG. 2 is a block diagram schematically illustrating a redundant excitation linear prediction (RELP) decoding apparatus corresponding to FIG. 1;

FIG. 3 (A) is a block diagram schematically showing a multi-pulse excitation linear predictive coding apparatus according to the present invention;

Fig. 3 (B) is a block diagram schematically showing a multi-pulse excitation linear prediction decoding apparatus corresponding to Fig. 3 (A)

4 (A) is a block diagram of a linear prediction (LPC) analysis filter shown in FIG. 3 (A)

4 (B) is a block diagram of the peak value and distance (PVD) encoder shown in FIG. 3 (A)

5 (A) to 5 (C) are signal waveforms for explaining the operation of the peak value and distance (PVD) encoder and decoder shown in FIG. 4 (B)

FIG. 6 is a specific block diagram of the peak value and distance (PVD) decoder shown in FIG. 3 (B).

BRIEF DESCRIPTION OF THE DRAWINGS

100: LPC analysis filter 100A: preprocessor

100B: LPC analyzer 100C: filter

200: PVD encoder 200A: Peak value calculator

200B: distance calculation unit 300: PVD decoder

300A: difference signal operation unit 300B: step size operation unit

300C: Signal waveform reproduction unit 300D: D / A conversion unit

300E: Low pass filter 400: LPC synthesis filter

A preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings.

3 (A) is a block diagram schematically illustrating a multi-pulse excitation linear prediction encoding apparatus according to the present invention.

As shown in the figure, the multi-pulse excitation linear prediction coding apparatus of the present invention comprises an LPC analysis filter 100 and a PVD coder 200.

The LPC analysis filter 100 receives the speech signal X (z) and obtains and outputs the LPC coefficient Ai and the prediction error E (z). As shown in FIG. 4 (A) A processor 100A, an LPC analyzer 100B and an inverse filter 100C.

Here, the preprocessor 100A removes noise components in a low frequency band through high-pass filtering on the input speech signal [X (z)], and performs a hamming windowing on the noise- And outputs it. Here, it is preferable that the high-pass filtering is cut off at about 70 Hz.

Then, the LPC analyzer 100B obtains a k parameter corresponding to the LPC coefficient through LPC analysis using, for example, a Schur recursion algorithm, quantizes the k parameter, and outputs an LPC coefficient Ai . The quantization may use scalar quantization or vector quantization.

The inverse filter 100C filters the speech signal input X (z) through the preprocessor 100A using the LPC coefficient Ai as a filter coefficient to output a prediction error E (z) do.

On the other hand, the PVD encoder 200 receives the prediction error [E (z)] from the LPC analysis filter 100 in units of one frame (for example, 10 msec) 4 (B), the peak value calculating section 200A calculates the peak error value E (z) by using the peak value and the distance algorithm PVDA to obtain the magnitude (amplitude) of the peak value, (N-1)], the current signal value E (n), and the next signal value E (n + 1) satisfy the following relationship (1) It is determined as the peak position, and the current signal value is determined as the maximum peak value. (K-2), t (k), and t (k + 2) correspond to the peak values of p (k- Corresponds to the position of the peak value.

[Equation 1]

E (n-1) < E (n) < E (n +

The peak value calculator 200A calculates the peak value E (n) and the current peak value E (n) for the input prediction error E (z) (n + 1)] is expressed by the following equation (2), the current signal value is determined as the minimum peak value, and the current position is determined as the peak position. (K-1), t (k + 1), and t (k + 1) correspond to the peak values of p k + 3) corresponds to the position of the peak value.

&Quot; (2) "

E (n-1) > E (n) > E (n +

Then, the distance calculator 200B calculates a distance between adjacent peak value positions at which the peak occurs. That is, as shown in Figs. 5A and 5B, t (k-2) and t (k-1), positions at which the peak values p (k- The distance d (2) can be obtained from t (k-1) - t (k-2). The distances d (3), d (4), and d (5) of the peak value positions can be obtained by the same method.

The PVD encoder 200 outputs the magnitude information and the distance information of the peak value obtained as described above as a coded predictive error signal C (E (z)) and outputs the encoded differential error signal C (E (z)) is channel-coded by a channel encoder (not shown) together with the above-described LPC coefficient [Ai] and output as a transmission channel.

3 (B) is a block diagram schematically showing a multi-pulse excitation linear prediction decoding apparatus according to the present invention.

As shown in the figure, the multi-pulse excitation linear prediction decoding apparatus of the present invention comprises a PVD decoder 300 and an LPC synthesis filter 400.

The PVD decoder 300 calculates a difference signal and a step size of neighboring peak values using the magnitude information and the distance information of the peak value and restores the linear prediction error signal using the difference signal and the step size. And includes a difference signal calculating unit 300A and a step size calculating unit 300B and a signal waveform reproducing unit 300C and a digital / analog converting unit 300D and a low pass filter 300E.

First, a coded signal transmitted through a channel is decoded by a channel decoder (not shown) and divided into peak information, size information and distance information as LPC coefficients Ai and encoding prediction errors [C (E))] Temporarily stored in a frame buffer (not shown), and sequentially output in frame units.

Here, the difference signal operation unit 300A calculates the difference signal of adjacent peak values using the size information of the peak value among the encoding prediction errors [C (E) z)) input through the frame buffer. That is, the difference signal calculation unit 300A can obtain the difference signal App (k) by the following equation (3).

&Quot; (3) "

p (1) - p (0) = App (1)

p (2) - p (1) = App (2)

p (3) - p (2) = App (3)

.

.

.

Then, the step size calculator 300B calculates the step size using the difference signal App (k) and the distance signal d (k). This step size [St (k)] can be obtained by the following equation (4).

&Quot; (4) "

App (1) ÷ d (1) = St (1)

App (2) ÷ d (2) = St (2)

App (3) ÷ d (3) = St (3)

.

.

.

Next, the signal waveform reproducing unit 300C calculates the signal waveform [App (k)] and the step size [St (k)] using the following equation (5) (k)].

&Quot; (5) "

INT (k) = p (k-1) + [d (k)

Here, INT (k) is the current reproduction waveform, p (k-1) is the size of the previous reproduction waveform, d (k) is the current distance signal, and St (k) is the current step size.

&Quot; (6) "

a = INT (1.1) = p (0)

b = INT (1.2) = INT (1.1) + St (1)

c = INT (1.3) = INT (1.2) + St (1)

d = INT (2.1) = p (1)

e = INT (2.2) = INT (2.1) + St (2)

f = INT (2.3) = INT (2.2) + St (2)

g = INT (2.4) = INT (2.3) + St (2)

h = INT (2.5) = INT (2.4) + St (2)

i = INT (2.6) = INT (2.5) + St (2)

j = INT (2.7) = INT (2.6) + St (2)

k = INT (3.1) = p (2)

l = INT (3.2) = INT (3.1) + St (3)

.

.

.

As shown in Fig. 5 (C), the calculation of the reproduction waveform INT (k) is performed for each peak value by the corresponding distance information. This is done by connecting p (k) and p The result is the same as the result of performing linear interpolation. In the present embodiment, for example, 80 reproduction waveforms [INT (k)] are created for one frame.

Then, the analog / digital conversion unit 300D converts the digital reproduction waveform [INT (k)] as shown in 5C into an analog signal.

Next, the low-pass filter 300E smoothes the analog signal by smoothing filtering to obtain a nearly smooth prediction error signal E (z); Interpolated excitation signal].

As described above, in the PVD decoder 300 described above, a retroreflection waveform approximates to the original signal is generated. Actually, the actual signal is not linear but nonlinear, but it is assumed that the actual signal is close to linear. Actually, the prediction error signal E (z) changes almost gently as shown in Fig. 5 (A).

In order to further interpolate the signal closer to the original signal, non-linear interpolation can be performed by changing the step value to a predetermined function so that it can be further approximated to the original signal.

The LPC synthesis filter 400 synthesizes the original speech signal X (z) with the LPC coefficient Ai through the frame buffer and the prediction error signal E (z) reconstructed by the PVD decoder 300 Output. Thereafter, filtering is performed through a post-processing filter (not shown) to restore a voice signal closer to the original signal.

As described above, in the present invention, a natural voice signal can be restored at a generated bit rate of 8 Kbps or less using a new hybrid coding / decoding scheme using a peak value and a distance algorithm (PVDA) that effectively encodes a prediction error signal have.

Next, the multi-pulse excitation linear prediction encoding apparatus according to the present invention can be implemented using one DSP (Digital Signal Providing) chip, and will be described by way of example. Here, since the main information of the input speech signal is distributed in a band of 4 KHz or less, it is assumed that a signal of 4 KHz or more is filtered and removed, and then the signal is sampled at 8 KHz and 80 samples are processed in one frame for 10 msec (each sample interval is 125 μsec) .

First, sample data obtained by A / D conversion of the input speech signal and sampled at 8 KHz is stored in two different memories (or different areas of the same memory) on a frame basis. Here, the DSP chip accesses the memory in which the sample data of the current frame is stored, reads the sample data, and calculates the LPC coefficient Ai and the error signal E (z) for 80 samples. At this time, Sample data is stored. Generally, it is known that the time required for calculating the LPC coefficients and the error signal in the DSP chip is 3MIPS (Million Instruction Per Second).

Therefore, when 10 MIPS DSP chip is used, 10 ⁵ instructions in 1 frame (10 msec) can be used, and 0.3 × 10 ⁵ instructions can be used for LPC calculation. Therefore, 0.7 × 10 ⁵ instructions can be allocated for the PVD encoding operation.

On the other hand, the error data [E (z)] calculated in the DSP chip is stored in two different memories (or different areas of the same memory) on a frame basis. Also, the DSP chip accesses the memory where the current frame is stored, reads the error data, and performs the PVD encoding operation for 80 samples. At this time, the error data of the next frame is stored in the other memory.

Here, the number of instructions required for an operation on one error data sample is calculated as follows. That is, since three samples must be sequentially compared in order to calculate the peak value, the sample input to the DSP chip is input to the register and shifted three times. At this time, three move commands are necessary. In order to compare three neighboring samples, two SUBTRACT commands are required for each of the equations (1) and (2).

Thereafter, the position of each sample data is counted by using one increment instruction, and when the peak value is obtained in the above operation, a subtraction instruction is required to calculate the distance from the previous peak value.

As described above, at least about 9 instructions are required for one sample for PVD, and about 720 instructions are required for one frame (80 samples), which is 0.7 × 10 ⁵ Quot; and " 0 " are assigned to the peak value and the distance value.

Therefore, the multi-pulse excitation linear prediction coding apparatus according to the present invention can sufficiently process LPC analysis and PVD algorithm in real time even by using a low-speed DSP chip of 10 MIPS.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

That is, for example, in the above-described embodiment, the error signal E (z) output from the LPC analysis filter 100 in the peak value calculation section 200A of the PVD coder 200 shown in FIG. However, in order to further magnify the transmission data, a threshold value is set for the peak value, a peak value within the threshold value is ignored, and a peak value that deviates from the threshold value range And only the position value may be output. At this time, the threshold value can be set by an iterative experiment in consideration of the amount of data to be transmitted and the sound quality.

As described above, according to the present invention, the reconstructed speech signal can be more approximated to the original signal waveform than that reproduced by the conventional RELP method.

In the encoding apparatus of the present invention, since data of all samples is encoded and transmitted in the conventional system, only the important sample information is transmitted, so that the amount of transmission information can be significantly reduced.

That is, since a natural voice signal can be restored at a generated bit rate of 8 Kbps or less, the present invention can reduce the generated bit rate to about 1/13 of that of a conventional audio signal, compared with the case of adopting the PCM method as a conventional waveform protector In case of a voice signal, it can be reduced by about 1/5. In addition, the present invention can reduce the generated bit rate to about 1/6 in case of a general audio signal, compared with the case of adopting the ADM method as a conventional waveform coder, Can be improved.

Claims (24)

An encoding apparatus for encoding a speech signal, comprising: A LPC analysis filter means for receiving a speech signal and obtaining and outputting a LPC coefficient and a prediction error signal; And a coding means for obtaining the magnitude of the peak value and the position of the peak value with respect to the prediction error signal from the LPC analysis filter, and outputting the obtained result. 2. The apparatus of claim 1, wherein the predictive-analysis filter means comprises: preprocessing means for removing noise components in the low-frequency band of the input speech signal and windowing the processed speech signal; Linear prediction analyzing means for obtaining a parameter through a linear prediction analysis of an output signal of the preprocessing means, quantizing the parameter and outputting the parameter as the linear prediction code coefficient, And filter means for filtering the speech signal using the LPC coefficient and outputting an error signal. 2. The apparatus of claim 1, wherein the encoding means comprises: peak value calculating means for receiving the error signal, computing a peak value of the signal and outputting the size information of the peak value and its position information; And a distance calculation means for calculating and outputting a distance between adjacent peaks based on the position information. 4. The multi-pulse excitation linear prediction encoding apparatus according to claim 3, wherein the peak value calculation means takes only peak values out of a predetermined threshold value for the calculated peak values. The apparatus of any one of claims 1 to 4, further comprising channel coding means for coding the position of the peak value and the magnitude of the peak value with respect to the linear prediction code coefficient and the error signal, and outputting the encoded signal through a transmission channel Wherein the multi-pulse excitation linear prediction encoding apparatus comprises: 3. The multi-pulse excitation linear prediction encoding apparatus of claim 2, wherein the low frequency band is 0 to 70 Hz, and the windowing process is a Hamming windowing process. 3. The apparatus of claim 2, wherein the linear prediction analysis is performed using a Shoe iteration algorithm. There is provided a decoding apparatus for receiving a linear predictive coding coefficient and a distance information between neighboring peak values for a prediction error signal and size information of a peak value as an encoded signal obtained by encoding a speech signal in an encoding apparatus and decoding the received information into an original speech signal , Decoding means for calculating a difference signal and a step size of adjacent peak values based on size information of the peak value and distance information and decoding the prediction error signal using the difference signal and the step size, And synthesis filter means for synthesizing the linear prediction code coefficient and a prediction error signal of the decoding means to output an original speech signal. The apparatus as claimed in claim 8, further comprising a preprocessing means for separating a transmission signal transmitted through a transmission channel into the linear prediction code coefficient, the distance information and the size information, Excitation linear prediction decoding apparatus. 10. The apparatus according to claim 8 or 9, wherein the decoding means comprises difference signal calculation means for calculating a difference signal of adjacent peak values based on the size information, A step size calculating means for calculating a step size using the distance information and a difference signal from the difference signal calculating means, A signal waveform reproducing means for reproducing a signal waveform by linear interpolation using the difference signal and the step size, Conversion means for converting the reproduction waveform of the signal waveform reproduction means into an analog signal, And a filtering means for performing low-pass filtering on the analog waveform signal from the conversion means. 11. The multi-pulse excitation linear prediction decoding apparatus according to claim 10, wherein the step size is obtained by dividing the difference signal by the distance information. 11. The apparatus according to claim 10, wherein the signal waveform reproducing means performs waveform reproduction by linear interpolation using the previous reproduction waveform size, current distance information, and current step size in units of distance information for each of the peak values Wherein the multi-pulse excitation linear predictive decoding apparatus is a multi-pulse excitation linear prediction decoding apparatus. A coding method for coding a speech signal, A LPC analysis filtering step of receiving a speech signal and obtaining a LPC coefficient and a prediction error signal, And a coding step of calculating a position of a peak value and a magnitude of a peak value with respect to a prediction error signal from the LPC analysis filtering step and outputting the result. 14. The method of claim 13, wherein the predictive code analysis filtering step comprises a preprocessing step of removing noise components of a low frequency band of the input speech signal and performing windowing processing, A linear prediction analysis step of obtaining a parameter through a linear prediction analysis of the output signal of the preprocessing step, quantizing the parameter and outputting it as a linear prediction code coefficient, And a filtering step of filtering the speech signal using the LPC coefficients and outputting a prediction error signal. 14. The method of claim 13, wherein the encoding step comprises: a peak value calculating step of receiving the prediction error signal, calculating a peak value of the signal, calculating size information of the peak value, And calculating a distance between adjacent peaks based on the position information and outputting the distance. 16. The multi-pulse excitation linear prediction encoding method according to claim 15, wherein the peak value calculation step takes only peak values out of a predetermined threshold value for the calculated peak values. The method of any one of claims 13 to 16, further comprising: encoding a position of a peak value and a magnitude of a peak value for the LPC code and the prediction error signal and outputting the encoded signal through a transmission channel Wherein the multi-pulse excitation linear prediction encoding method comprises: 15. The method of claim 14, wherein the low frequency band is 0 to 70 Hz, and the windowing process is a Hamming windowing process. 15. The method of claim 14, wherein the linear prediction analysis is performed using a Shoe iteration algorithm. A decoding method for receiving a linear predictive coding coefficient and a distance information between neighboring peak values of a predictive error signal and size information of a peak value as an encoded signal obtained by encoding a speech signal in an encoding apparatus and decoding the received information into an original speech signal, Calculating a difference signal and a step size of neighboring pig values based on size information of the peak value and distance information, and decoding the prediction error signal using the difference signal and the step size, And a synthesis filtering step of synthesizing the linear prediction code coefficient and the prediction error signal of the decoding step and outputting the original speech signal. 21. The method of claim 20, further comprising a preprocessing step of separating a transmission signal transmitted through a transmission channel into the LSP coefficient, the distance information, and the size information, A linear predictive decoding method. 22. The method of claim 20 or 21, wherein the decoding step comprises: a difference signal calculation step of calculating a difference signal of adjacent peak values based on the size information; A step size calculation step of calculating a step size using the distance information and the difference signal from the difference signal calculation step, A signal waveform reproducing step of reproducing a signal waveform by linear interpolation using the difference signal and the step size, A conversion step of converting the reproduction waveform of the signal waveform reconstruction step into an analog signal, And performing a low-pass filtering on the analog waveform signal from the conversion step. 23. The multi-pulse excitation linear prediction decoding method according to claim 22, wherein the step size is obtained by dividing the difference signal by the distance information. The method according to claim 22, wherein the signal waveform reproducing step performs waveform reproduction using linear interpolation using the previous reproduction waveform size, current distance information, and current step size in units of distance information for each of the peak values Wherein the multi-pulse excitation linear predictive decoding method comprises: