JP3088204B2 - Code-excited linear prediction encoding device and decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is suitable for application to a code-excited linear predictive coding apparatus and a decoding apparatus which comply with a code-excited linear predictive coding method (CELP coding method).

[0002]

2. Description of the Related Art The CELP coding method proposed by Atal et al. Is an effective method as a high-efficiency coding method for audio signals at a compression rate of 8 kbit / s or less.

[0003] This belongs to a frame encoding method for encoding a speech signal for each frame of a fixed length, and includes a vocal tract parameter required once for each frame and a subframe obtained by decomposing the frame into several subframes. It is expressed by the parameters of the excitation source obtained for each. In a forward-type CELP coding method for obtaining vocal tract parameters from input speech, two parameters are used as excitation source parameters: a statistical codebook for coding unvoiced sounds and an adaptive codebook for coding voiced sounds. In many cases, vector quantization is performed, for example, as described in the following document.
A speech signal can be considered as being divided into a stationary voiced section and a transient unvoiced section, and both have significantly different statistical properties. Therefore, as described above,
A statistical codebook and an adaptive codebook are separately prepared.

Reference: NSJayant & JHChen, "Speech Co
ding with Time-Varying BitAllocations to Excitatio
n and LPC parameters ", Proc, ICASSP-89, (1989) Fig. 2 illustrates the concept of coding using an adaptive codebook. The sound quality of a voiced section having periodicity (autocorrelation) is shown. In the adaptive codebook having a large contribution, an excitation source waveform S1 for a long time T immediately before the current subframe is stored, for example, by a shift register. Consider partial waveforms (statistical excitation code vectors) S11 to S1x having a sub-frame period.If any of partial waveforms S11, S12,... , Searching for whether the input sound and the locally reproduced sound are the closest, and specifying the searched partial waveform S1n (eg, the first partial waveform S1n from the first sample of the long time T). The number of samples up to the sample is expressed in bits) as an output code for the current processing target subframe from the adaptive codebook, where the output code is a waveform substantially similar to the current processing target subframe. It is a time difference (lag: pitch cycle) from a past time point having a shape.

[0005]

Since the voiced sound to which the adaptive codebook greatly contributes has a high stationarity, the lag (pitch cycle) fluctuates slowly and the lag between adjacent subframes is similar. Become.

However, conventionally, such a property of the lag is not used, and the lag is independently obtained for each subframe. That is, as described above, for each subframe, many partial waveforms S11, S12,
... The optimum lag was determined from S1x to determine the lag. Therefore, much processing and time were required to obtain a lag.

Further, as described above, without utilizing the property that the lag between successive subframes is similar,
Since the lag obtained for each subframe is transmitted, it can be said that redundant information is included in the transmission information. The CELP coding method is often adopted as low bit rate voice coding, and it is required to effectively use each bit of transmission information. In this regard, coding for lag can be considered to be insufficient. .

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and has a code excitation linear predictive coding which can reduce the processing and time required to determine the sign of excitation source information contributing to voiced sound. It was intended to provide a container.

Another object of the present invention is to provide a code excitation linear prediction encoder and a decoder capable of reducing the number of bits required for encoding excitation source information contributing to voiced sound as compared with the prior art. is there.

[0010]

According to the present invention, there is provided a code excitation linear predictive coding apparatus having an adaptive codebook, comprising: a coding period of excitation source information using an adaptive codebook; integer multiple of the period der
Includes a long-period analyzer for determining the pitch period of the input speech signal for each long period that, each coding cycle in the long period of a plurality of adaptive excitation code vectors stored in the adaptive codebook
When searching for the optimal adaptive excitation code vector for each, the optimal adaptive excitation code for each coding cycle
Shift vector by pitch period for each encoding period
The search is performed from a predetermined number of adaptive excitation code vectors before and after the adaptive excitation code vector at the time position .

Here, the coded speech transmitted to the code-excitation linear predictive decoding apparatus has a pitch cycle corresponding to every long cycle, and a pitch cycle corresponding to every coding cycle in the long cycle. Coding
It is preferable to include each time difference information between the optimal adaptive excitation code vector obtained as a result of the search for the period and the adaptive excitation code vector at a time position shifted by the pitch period .

[0012] In the decoding apparatus corresponding to the code excitation linear predictive encoding apparatus of such a preferred embodiment , each of the encoding in the pitch period and the long period included in the encoded speech is performed .
Each value obtained by adding each time difference information corresponding to the cycle ,
Each as an index into the adaptive codebook
It is preferable to output an adaptive excitation code vector corresponding to each encoding cycle .

[0013]

According to the present invention, the pitch period of a voiced sound section is
Obtained in a long cycle that is an integral multiple of the information encoding cycle
Are also based on the fact that they take almost the same value even when they are obtained in the coding cycle .

By using this property, it is intended to reduce the search time and processing of the adaptive codebook. In the code excitation linear prediction encoding device according to the present invention, the long-period analyzer has a period that is an integral multiple of the encoding period (search period) of the excitation source information using the adaptive codebook.
For each long period , the pitch period of the input speech signal is determined, and the code for the adaptive codebook is determined by using a plurality of adaptive excitation code vectors stored in the adaptive codebook from each of the encoding periods in the long period. When searching for the optimal adaptive excitation code vector for
The optimal adaptive excitation code vector for each encoding cycle.
The search is made from a predetermined number of adaptive excitation code vectors before and after the adaptive excitation code vector at the time position shifted by the pitch period from the period .

Here, the coded speech transmitted to the code-excitation linear predictive decoding apparatus has a pitch cycle corresponding to every long cycle, and a pitch cycle corresponding to every coding cycle in the long cycle. Coding
By including each time difference information between the optimal adaptive excitation code vector obtained as a result of the search for the period and the adaptive excitation code vector at the time position shifted by the pitch period , the compression efficiency can be increased compared to the conventional case. It is possible and preferable.

[0016] In the decoding apparatus corresponding to the code-excitation linear predictive coding apparatus of such a preferable embodiment, since information for directly accessing the adaptive codebook is not transmitted, the pitch period included in the coded speech is not determined. Long circumference
Each value obtained by adding the respective time difference information corresponding to each coding cycle during the period will access the adaptive excitation code vector corresponding to each coding cycle as an index to each adaptive codebook.

[0017]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a linear excitation coding apparatus and a decoding apparatus according to an embodiment of the present invention;

FIG. 1 shows the configuration of a code excitation linear prediction encoding apparatus according to an embodiment.

In FIG. 1, an input audio signal sequence that has been subjected to analog / digital conversion is input to the encoding apparatus in units of a specific frame length.

The vocal tract analyzer 11 performs vocal tract analysis (for example, LPC analysis) of the input speech signal, and converts the vocal tract parameter α obtained by the analysis into a long-period analyzer 12, an excitation source evaluation circuit 13, and a multiplexer 17 Give to. Here, the vocal tract parameter α is obtained for each frame and output for each frame.

The long-period analyzer 12 calculates a long-period correlation of the input voice signal, obtains a lag (hereinafter, referred to as a frame lag) L which is pitch period information for the entire frame of the input voice signal, and Evaluation circuit 13 and multiplexer 17
Give to. Such a frame lug L is also 1 in each frame.
It is asked every time. The long-period analyzer 12 includes:
A sound path parameter α, which is a short-period parameter, is also given. Using this sound path parameter α, the influence of a short-period component in the input speech signal is removed to detect a long-period correlation.

The notation method and detection accuracy of the frame lag L obtained by the long period analyzer 12 will be described later.

The adaptive code book 14 stores a plurality of adaptive excitation code vectors that greatly contribute to a voiced sound section, and includes, for example, a shift register and a table. As will be described later, the optimum adaptive excitation code vector is obtained for each sub-frame obtained by equally dividing one frame period under the control of the excitation source evaluation circuit 13.

The statistical code book 15 stores a plurality of statistical excitation code vectors that greatly contribute to an unvoiced sound section. As will be described later, the optimum statistical excitation code vector is also obtained for each subframe under the control of the excitation source evaluation circuit 13.

The gain codebook 16 stores a plurality of pieces of gain information for each of the adaptive excitation code vector and the statistical excitation code vector. As described below,
The optimal gain (β, γ) for the optimal adaptive excitation code vector and the statistical excitation code vector is also obtained for each subframe under the control of the excitation source evaluation circuit 13.

The excitation source evaluation circuit 13 determines excitation source information ΔL, I, (β, γ) excluding the frame lag L for each subframe. The excitation source evaluation circuit 13 includes excitation source information (hereinafter, referred to as a subframe correction lag) ΔL relating to the adaptive codebook 14 and excitation source information (hereinafter, referred to as a statistical code index) relating to the statistical codebook 15.
I, excitation source information related to the gain codebook 16 (hereinafter, referred to as
The determination operation is performed in the order of (β, γ).

The excitation source evaluation circuit 13 has a configuration for obtaining a local reproduction voice from the excitation code vector and the vocal tract parameter α, and a configuration for calculating the similarity between the local reproduction voice and the input voice (for example, (See Japanese Patent Application No. 3-327443, drawings, etc.), and search for information that maximizes the similarity in the adaptive codebook 14, statistical codebook 15, gain codebook 16, and the like.

The multiplexor 17 includes a vocal tract parameter α and a frame lag L given once to a frame, a subframe correction lag ΔL given once to a subframe, a statistical code index I and an optimum gain (β, γ). Are multiplexed while appropriately quantizing to obtain an encoded audio signal.

The code-excited linear prediction encoding apparatus according to the embodiment having the above-described configuration operates as follows.

When an input speech signal for one frame is given, the speech analyzer 11 and the long-period analyzer 12 respectively
The vocal tract parameter α and the frame lag L are obtained and analyzed and output.

The excitation source evaluation circuit 13 determines excitation source information for each subframe. First, the excitation source evaluation circuit 13
With the output from the statistical codebook 15 and the gain codebook 16 stopped (or with the default value output), a plurality of adaptive excitation codes from the adaptive codebook 14 as candidates for the optimal adaptive excitation code vector. Vectors are output, for example, time-sequentially or in parallel, the respective local reproduction sounds are obtained and evaluated using the adaptive excitation code vector of each candidate, and the local reproduction sound having the highest similarity to the input sound signal is synthesized. The obtained adaptive excitation code vector is set as an optimal adaptive excitation code vector.

Here, the excitation source evaluating circuit 13 sets a predetermined number of adaptive excitation code vectors before and after the adaptive excitation code vector at the time position shifted from the current subframe by the frame lag L as candidates. That is, unlike the related art, all adaptive excitation code vectors stored in the adaptive codebook 14 are not set as candidates.
The method for determining the candidate vector will be described later.

Upon determining the optimal adaptive excitation code vector, the excitation source evaluation circuit 13 determines the optimal adaptive excitation code vector at the time position shifted by the frame lag L from the current subframe and the optimal adaptive excitation code vector. Multiplexer 1 obtains time difference information and sets it as subframe correction lag ΔL
7 is output.

Next, the excitation source evaluation circuit 13 stops the output from the gain codebook 16 (or outputs the default value), and outputs the statistical codebook 15.
All statistical excitation code vectors are output as candidates for the optimal statistical excitation code vector from, for example, time-sequentially or in parallel, and the statistical excitation code vector of each candidate and the optimal adaptive excitation code vector are added and obtained by addition. Using the obtained excitation code vectors, a local reproduction audio is obtained and evaluated, and a statistical excitation code vector obtained by synthesizing a local reproduction audio having the highest similarity to the input audio signal is set as an optimal statistical excitation code vector. . Then, an index (statistical code index) I for the optimal statistical excitation code vector is output to the multiplexer 17.

Subsequently, the excitation source evaluation circuit 13 causes the gain codebook 16 to output all the gain information as candidates for the optimality gain, for example, in time sequence or in parallel, and optimizes the statistical excitation based on the gain information of each candidate. Weighted addition of the code vector and the optimal adaptive excitation code vector,
Using each excitation code vector obtained by the addition, a locally reproduced sound is obtained and evaluated, and gain information obtained by synthesizing the locally reproduced sound having the highest similarity to the input sound signal is determined as the optimum gain. In the case of this embodiment, the optimum gain (β, γ) itself is given to the multiplexer 17.

As described above, when the optimal adaptive excitation code vector, statistical excitation code vector, and gain for the current subframe are obtained, the final excitation code vector (voice voice) for the current subframe is obtained from these information. An excitation code vector (combined with the path parameter α) is obtained and given to the adaptive codebook 14. The adaptive codebook 14 takes this excitation code vector as the latest one, and discards the oldest adaptive excitation code vector.

When such adaptive processing is completed, the process proceeds to the processing of the next subframe.

Then, the multiplexer 17 gives the vocal tract parameter α and the frame lag L given once to the frame, the subframe correction lag ΔL given once to the subframe, the statistical code index I, and the optimality gain (β, The encoded audio signal is assembled from γ) and output to the communication path.

FIG. 3 is an explanatory diagram of the notation method (detection accuracy) of the frame lag L obtained by the long period analyzer 12. FIG. 3 shows three types of notation method and detection accuracy of the frame lag L, and any of them can be used in practice.

The pitch frequency of voice is from 50 Hz to 400
Hz, so for example, if the input audio signal is 8000 Hz
When the analog / digital conversion is performed, the value of the frame lag L is set to 150 (a value close to 8000/50) from the sampling cycle of 20 (= 8000/400).
It is appropriate to set the sampling period to a range of values that can be taken. In the case of this embodiment, since the frame lag L is naturally processed as a digital value, it is expressed in bits.

FIG. 3A shows a first notation method of the frame lag L. In consideration of the bit representation, the range of the value of the frame lag L is set to 20 sampling periods to 147 sampling periods, and all values (128 values) different by one sampling period are taken. Therefore, in this case, the frame lag L is represented by 7 bits.

FIG. 3B shows a second notation method of the frame lag L. Considering the bit representation, the range of the value of the frame lag L is set to 20 sampling periods to 146 sampling periods, and all values (64 values) differing by two sampling periods are assumed. Therefore, in this case, the frame lag L is represented by 6 bits. Since the frame lag L gives an indication of the time position where the optimal adaptive code vector exists in the adaptive code book 14, even if the frame lag L is described so as to differ by two sampling periods (even if the precision is lowered). There is no problem.

FIG. 3C shows a third notation method of the frame lag L. In this method, the minimum value of the frame lag L is set to 20 sampling periods, the value of the adjacent frame lag L is determined at intervals of one sampling period in the portion corresponding to the high pitch frequency, and the portion corresponding to the low pitch frequency is determined. , The values of the adjacent frame lags L are determined at wider intervals. In this case, when the frame lag L is expressed with a small number of bits, it can be efficiently expressed.

FIG. 4 explains the search range of the optimal adaptive excitation code vector and the method of expressing the subframe correction lag ΔL, and shows four types of methods.

[0045] FIG. 4 (a) is an explanatory view of the first method, the adaptive excitation code the samples prior to only 8 sampling periods from the beginning samples of the adaptive excitation code vector corresponding to full Remuragu L and the top sample vector From
A total of 1 from the first sample of the adaptive excitation code vector corresponding to the frame lag L to the adaptive excitation code vector whose first sample is a sample after 7 sampling periods.
This shows that six adaptive excitation code vectors are used as candidate vectors. Therefore, the subframe correction lag ΔL can be expressed by 4 bits.

FIG. 4B is an explanatory diagram of the second method, in which the adaptive excitation code vector whose first sample is a sample eight sampling periods before the first sample of the adaptive excitation code vector corresponding to the frame lag L. From
The adaptive excitation code vector, which is 7.5 seconds after the first sample of the adaptive excitation code vector corresponding to the frame lag L to the adaptive excitation code vector having the first sample, This shows that a total of 32 adaptive excitation code vectors having a time difference of 0.5 sampling period are set as candidate vectors. Therefore, the subframe correction lag ΔL can be represented by 5 bits. This is a method when an upsampling filter is used for the adaptive codebook 14.

FIG. 4C is an explanatory diagram of the third method.
This method also assumes that an upsampling filter is used for the adaptive codebook 14, and in the vicinity of the adaptive excitation code vector corresponding to the frame lag L,
Consider an adaptive excitation code vector which is a candidate by shifting the position of the first sample by a time shorter than one sampling period, and as the distance from the adaptive excitation code vector corresponding to the frame lag L increases, the time difference between the adaptive excitation code vectors of adjacent candidates. Is made larger. Therefore, the subframe correction lag ΔL can be efficiently expressed.

FIG. 4D is an explanatory diagram of the fourth method.
This method reduces the time difference between the adaptive excitation code vectors of adjacent candidates in a range where the frame lag L is small (range where the pitch frequency is high), and reduces the time difference in a range where the frame lag L is large (range where the pitch frequency is low). The time difference between adaptive excitation code vectors of matching candidates is set to be large. Therefore, even if the subframe correction lag ΔL is represented by a predetermined number of bits, the value of one bit differs depending on the value of the frame lag L.

As described above, in this embodiment, the search range of the adaptive codebook 14 is set by obtaining the frame lag L. This is for the following reason. The fluctuation of the pitch period in the voiced sound section is slow, and the values are similar whether the pitch period (frame lag L) is obtained for a frame or the pitch period (frame lag L + subframe correction lag ΔL) is obtained for a subframe. It has become. Therefore, even if the pitch period of the frame (frame lag L) is used as the predicted value of the pitch period of the subframe, and the pitch period of the subframe is searched before and after that, the accurate pitch period of the subframe can be obtained. Thus, the search time and the amount of search processing can be reduced.

Further, in this embodiment, the coded voice signal is transmitted including the frame lag L and the subframe correction lag ΔL for the following reason. Conventionally, the pitch period (frame lag L +
(Corresponding to the subframe correction lag ΔL). If one frame is composed of 4 subframes and the pitch period information is composed of 7 bits, 2 frames per frame are conventionally used.
8 bits will be transmitted. On the other hand, assuming that the subframe correction lag ΔL is 4 bits, in this embodiment, the pitch information (frame lag L) for the frame is once for one frame, and the pitch information (subframe correction lag ΔL) for the subframe. Is transmitted four times,
The number of bits of pitch information transmitted in one frame is 7 + 4
× 4 = 23 bits. That is, the compression efficiency can be increased as compared with the related art.

Next, the configuration and operation of the code-excited linear prediction decoding apparatus according to the embodiment will be described in detail with reference to FIG.

The coded speech signal transmitted from the code-excited linear predictive coding apparatus (FIG. 1) is supplied to the demultiplexer 20, where the vocal tract parameter α, the frame lag L, the subframe correction lag ΔL, the statistical code The index I is separated into the optimality gain (β, γ). The separated vocal tract parameter α is provided to the synthesis filter 26, and the separated frame lag L and subframe correction lag ΔL are added in the demultiplexer 20 and provided to the adaptive codebook 21, and the separated statistical code The index I is provided to the statistical codebook 22, and the separated optimality gain (β,
The gain β for the adaptive excitation code vector of γ) is given to the multiplier 23, and the gain γ for the other statistical excitation code vector is given to the multiplier 25 of γ).

Thus, the adaptive adaptive excitation code vector for the subframe is output from the adaptive codebook 21, multiplied by β by the multiplier 23 and input to the adder 24, and the subframe is output from the statistical codebook 22. Is output, and this is multiplied by γ by the multiplier 25 and input to the adder 24. Thereby, in the adder 24,
Finally, an excitation code vector for this subframe is obtained, and the synthesis filter 26 synthesizes the excitation code vector by applying the vocal tract parameter α to obtain a reproduced audio signal corresponding to the input audio signal. Adder 24
Are fed back to the adaptive codebook 21, and the adaptive codevectors in the adaptive codebook 21 are adaptively changed.

Therefore, according to the above embodiment, the adaptive code book 14 is determined by the frame lag L obtained for each frame.
, The coded audio signal is transmitted together with the frame lag L and the subframe correction lag ΔL, and the decoding device converts the frame lag L and the subframe correction lag ΔL from the subframe lag (for the subframe). Since the adaptive codebook 21 is accessed by obtaining (pitch period) L + ΔL, the processing and time for determining the sign of the excitation source information contributing to the voiced sound can be reduced as compared with the conventional art, and the contribution to the voiced sound can be reduced. The number of bits required for the sign of the excitation source information to be performed can be reduced as compared with the related art.

The present invention can be applied to any code-excited linear prediction encoding device and decoding device having an adaptive codebook, and is not limited to the forward type.

In the above embodiment, the long-period correlation value is obtained for each frame which is the analysis period of the vocal tract parameters. However, the present invention is not limited to this.
The analysis cycle of the vocal tract parameters and the cycle for obtaining the long-period correlation value may be different. However, it is necessary that the cycle for obtaining the long-period correlation value is an integral multiple of the search cycle of the adaptive codebook.

In the above embodiment, the frame lag L and the sub-frame correction lag ΔL are transmitted while being included in the coded audio signal. However, the frame lag L and the sub-frame correction lag ΔL are added. Frame lug L +
ΔL may be included in the encoded audio signal and transmitted. In this case, the configuration of the code-excited linear predictive decoding device is the same as the conventional one. In this case, the effect of increasing the compression efficiency of the excitation source information cannot be obtained as compared with the related art, but the effect of reducing the search time and processing of the adaptive codebook 14 can be obtained.

[0058]

As described above, according to the first aspect of the present invention, the pitch of the input audio signal is changed every long period which is an integral multiple of the encoding period of the excitation source information using the adaptive codebook.
It includes a long-period analyzer for determining the switch period, the length of a plurality of adaptive excitation code vectors stored in the adaptive codebook
When searching for the optimal adaptive excitation code vector for each coding cycle in the cycle,
The optimal adaptive excitation code vector is calculated for each encoding cycle.
The search is performed from a predetermined number of adaptive excitation code vectors before and after the adaptive excitation code vector at the time position shifted by the pitch period , so that the search time and processing for the adaptive codebook can be reduced. An excitation linear prediction encoding device can be realized.

According to the second aspect of the present invention, the coded speech transmitted to the code-excited linear predictive decoding apparatus is further provided with a pitch cycle for every long cycle and all the codes in the long cycle.
The optimal adaptive excitation code vector and pitch period obtained as a result of the search for the encoding period corresponding to the encoding period
Since each time difference information from the adaptive excitation code vector at the displaced time position is included, it is possible to realize a code excitation linear prediction encoding device that can also obtain the effect of increasing the compression efficiency.

According to a third aspect of the present invention, there is provided a code excitation linear prediction decoding apparatus for decoding coded speech transmitted from the code excitation linear prediction encoding apparatus according to the second aspect of the present invention. The pitch period included in the voice
Each value obtained by adding the respective time difference information corresponding to each coding cycle in the long period, since so as to output an adaptive excitation code vector corresponding to each coding cycle as an index to each adaptive codebook Thus, the effect according to the second aspect of the present invention can be made effective.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a code excitation linear prediction encoding device according to an embodiment.

FIG. 2 is an explanatory diagram of a concept of an adaptive code book.

FIG. 3 is a notation method (detection accuracy) of a frame lag according to the embodiment.
FIG.

FIG. 4 is an explanatory diagram of a search range of an adaptive codebook and a subframe correction lag according to the embodiment.

FIG. 5 is a block diagram illustrating a configuration of a code excitation linear prediction decoding device according to an embodiment.

[Explanation of symbols]

12: long-period analyzer, 13: excitation source evaluation circuit, 14: adaptive codebook, 20: demultiplexer, L: frame lag (long-period correlation value), ΔL: subframe correction lag (optimal adaptive excitation code vector (Time difference information between the adaptive excitation code vector corresponding to the long-period correlation value).

Continuation of the front page (72) Inventor Katsutoshi Ito 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-4-270398 (JP, A) JP-A-1- 179100 (JP, A) JP-A-5-289700 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-21/06 H03M 7/30 H04B 14/04 JICST File (JOIS)

Claims (3)

(57) [Claims] 1. A code excitation linear predictive coding apparatus provided with an adaptive codebook, wherein the pitch of the input speech signal is changed every long period that is an integral multiple of the encoding period of the excitation source information using the adaptive codebook.
Includes a long-period analyzer for determining the switch period, the coding cycle in the long period of a plurality of adaptive excitation code vectors stored in said adaptive codebook it
Search for optimal adaptive excitation code vector
The optimal adaptive excitation code vector for each encoding cycle.
To the encoding period by the pitch period described above.
A code excitation linear predictive coding apparatus characterized in that a search is performed from a predetermined number of adaptive excitation code vectors before and after the adaptive excitation code vector at the time position . 2. The coded speech transmitted to the code-excited linear predictive decoding device includes, for each of the long periods, the pitch period and the code corresponding to all the coding periods in the long period.
A time difference information between the optimal adaptive excitation code vector obtained as a result of the search for the encoding period and the adaptive excitation code vector at a time position shifted by the pitch period. Item 2. A code excitation linear prediction encoding device according to item 1. 3. A code-excited linear prediction decoding device for decoding encoded speech transmitted from the code-excited linear prediction encoding device according to claim 2, wherein the pitch period included in the encoded speech is And the above long period
Code each value obtained by adding the respective time difference information, and characterized by outputting the adaptive excitation code vector corresponding to each coding cycle as an index to each adaptive codebook corresponding to each coding cycle in Excitation linear prediction decoding device.