JPH08101700A - Vector quantization device - Google Patents

Abstract

PURPOSE: To provide a vector quantization device capable of suppressing sudden quality deterioration of a regenerative signal to the minimum even when a code error exists in a transmission line. CONSTITUTION: In the vector quantization device expressing a target vector by using a code vector specified by an index, a distortion evaluation part 2310 performs distortion evaluation of the code vector considering no code error of the index and the distortion evaluation considering the code error, and after a first selection part 2320 selects a small number of index candidates from many index candidates based on the distortion evaluation result considering no code error, a second selection part 2325 selects the index used for expressing the target vector based on the distortion evaluation considering the code error from a small number of index candidates selected by the first selection part 2320.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantizer used for low bit rate coding of voice and image information.

[0002]

2. Description of the Related Art A vector quantizer expresses a target vector using a code vector designated by an index, and is often used in a voice coder for compressing and coding voice for wireless transmission. .

CELP (Code Exc) is a voice encoding system capable of reproducing a relatively high quality voice even at a low bit rate.
Ited Linear Prediction) method is known. CEL
For details of the P method, see, for example, "MR Schroeder and Atal," Co
de-Excited Linear Prediction (CELP): high quality s
peech at very low bit rates ", in Proc. ICASSP'85,
pp.937-939,1985 "(Reference 1).

In the speech coding method represented by the CELP method in recent years, there are two types of adaptive codebooks and noise codebooks.
By expressing the pitch component shape and the noise component shape of the driving signal as the sound source by vector quantization using two codebooks, the driving signal obtained by combining these shapes is passed through a time-varying synthesis filter. By doing so, synthetic speech is generated. Further, in order to improve the quality of the synthesized speech, the driving signal coding unit performs coding so as to minimize the subjective distortion of the synthesized speech. That is, a method of selecting indexes for designating code vectors to be extracted from the adaptive codebook and the noise codebook by using a time-varying distortion measure extracted from speech is adopted. Therefore, conventional CELP
The method can provide comparatively high quality synthesized speech even at a low bit rate of about 4 kbit / sec.

However, when a conventional CELP system or the like is used in a voice coding device for mobile communication using radio, it is known that a code error in a transmission line greatly deteriorates the quality of synthesized voice. There is. Even if an error correction code is added, such a code error cannot be prevented if the state of the transmission line is bad. Therefore, it is important to provide a practical coding apparatus by incorporating a mechanism for minimizing the quality deterioration due to the code error of the transmission line into the speech coding process.

[0006] As a code error countermeasure method conventionally known in a speech coder, redundancy is provided in advance between transmission parameter candidates which are coded outputs, and a code vector and a code vector are set so that deterioration due to a code error is reduced. There is a method of taking measures such as presetting the association of indexes at the stage of designing an encoding device. These methods are described in detail in, for example, the Institute of Electronics, Information and Communication Engineers, vol.J77-A No.3 pp.485-493, 1994, “Moriya et al.“ Learning method for excitation codebook in CELP coding ”(Reference 2). Has been. When a coding device designed using such a method is used, when the code of the index information selected by the coding device receives a code error in the transmission line, the quality of the code vector reproduced by the decoding device. There is an effect that deterioration can be suppressed on average.

FIG. 5 is a diagram showing a configuration of a conventional speech coding apparatus, which corresponds to an adaptive codebook 200 which stores past drive signals and generates a code vector based on a designated pitch period, and a noise index. The noise codebook 218 storing the coded vectors, and the gain circuit 21 that gives a gain to each of the codevectors obtained from these codebooks 200 and 218 to generate a drive signal.
6, 225 and an adder 226, a synthesis filter 227 that obtains a synthesized voice by inputting a drive signal, a synthesis filter information analysis unit 229 that analyzes the filter information of the synthesis filter 227 from the input voice, and a target vector from the input voice and the synthesis filter information. The target vector creation unit 230 that generates the target vector and the synthesis filter information
A distortion evaluation unit 231 that obtains a distortion evaluation value for the synthesized voice obtained in step 27, and a code selection unit 232.

However, in the above-described conventional speech coding apparatus, at the stage of index selection performed by the coding apparatus during actual operation, the influence of quality deterioration due to a code error in the transmission path is not taken into consideration, and the distortion evaluation unit is not considered. In 231 the distortion of the code vector when there is no code error is evaluated, and the code of the noise index is selected by the code selection unit 232 only from the evaluation value. That is, the degree of distortion when a code error occurs in the selected index is not evaluated. Therefore, when a code error occurs in the index, there is a problem that a large distortion may occur in the code vector due to the code error, and a large quality deterioration of the reproduced signal may suddenly occur.

[0009]

As described above, in the conventional vector quantizer, when there is a code error in the code of the selected index, there is a possibility that the reproduced signal suddenly deteriorates in quality. There was a problem.
It is an object of the present invention to provide a vector quantizer which can minimize sudden quality deterioration of a reproduced signal even when there is a code error in a transmission line.

[0010]

In order to solve the above-mentioned problems, the present invention is a vector quantizer which expresses a target vector using a code vector specified by an index, and considers a code error in the index. Distortion evaluating means for evaluating the distortion of the vector, and selection means for selecting at least one index from a plurality of index candidates that can be indexes used for expressing the target vector based on the evaluation result of this distortion evaluating means It is characterized by

The present invention further includes a first distortion evaluating means for evaluating the distortion of the code vector, a second distortion evaluating means for evaluating the distortion of the code vector in consideration of a code error in the index, and a first distortion. Based on the evaluation result of the second distortion evaluation means, based on the evaluation result of the evaluation means, first selection means for selecting a small number of index candidates from a plurality of index candidates that can be indexes used to express the target vector. And second selection means for selecting at least one index from the index candidates selected by the first selection means.

Furthermore, the present invention considers the code error in the second selecting means based on the information detected by the detecting means for detecting the information regarding the code error of the transmission path for transmitting the information of the index. And means for adjusting the degree.

The distortion evaluating means in the present invention calculates and evaluates the distortion of the code vector. As a method of evaluating the distortion, besides the method of actually calculating the distance distortion between the vectors, A simple distortion evaluation method that combines a value equivalent to the inner product of the composite vector and the target vector by transformation, a value equivalent to the power of the composite vector or a value similar to this, or the direct shape of the code vector due to a code error. It is also possible to use such a method as to obtain the distortion of (1) without using a synthesis filter.

The present invention represents a speech coding apparatus using vector quantization, that is, synthesized speech is represented by a synthesis filter and a driving signal for driving the speech, and a noise component of the driving signal is represented by a noise code vector designated by an index. When applied to a speech coding apparatus based on the method of expressing by the method, the index selection method according to the present invention is applied in the process of selecting an index used for expressing the noise component of the drive signal from a large number of index candidates. It can be applied. In that case, weighted distortion evaluation is performed on the code vector.

[0015]

In the vector quantizer according to the present invention, a desired index is selected from a large number of index candidates that can be used as an index for expressing the target vector, based on the result of distortion evaluation of the code vector considering the code error of the index. By selecting, even if there is a code error in the code of the selected index, the possibility that a large quality deterioration of the reproduced signal will suddenly occur is greatly reduced.

Further, in another vector quantizer according to the present invention, by selecting a small number of index candidates from a large number of index candidates based on the code vector distortion evaluation result, the code is selected among the entire index candidates. A small number of index candidates are selected that can secure a certain quality or more with little distortion in the reproduced signal when there is no error. And then, for this few index candidates,
Distortion evaluation that considers the effect of index information bit errors is performed, and the number of index candidate candidates is further narrowed down based on the result of the distortion evaluation, so that index candidates with less quality deterioration at the time of code errors are selected. Select the index to be used to represent the target vector.

Usually, the distortion evaluation considering the influence of the code error complicates the calculation, but in the present invention, the distortion evaluation considering the error is performed after reducing the number of index candidates by the distortion evaluation not considering the code error. Therefore, it is possible to select an index that can stably suppress quality deterioration with respect to a code error in the transmission path, without increasing the calculation amount.

Further, when the present invention is applied to a coding apparatus or a wireless communication system having a mechanism for obtaining information about a code error on a transmission line, a code error is generated according to the information about the code error on the transmission line. By changing the degree of considering the effect of code error in the distortion evaluation in consideration of, by adapting to the code error state of the communication path, it becomes possible to select the index that can obtain a reproduced signal with less distortion, Furthermore, it is possible to perform vector quantization that is resistant to code errors in the transmission path.

Furthermore, since the present invention can provide a coding device that is resistant to code errors only by changing the code search method on the coding side, when applied to an already standardized coding system, a table such as a codebook is used. There is an advantage that there is no need to rewrite.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. (First Embodiment) FIG. 1 is a block diagram of a speech coder according to a first embodiment to which a vector quantizer according to the present invention is applied. The present embodiment has a configuration for extracting / encoding four types of information, that is, synthesis filter information, pitch period information, noise index information, and gain information, from input speech.
And, the pitch period information, the noise index information and the gain information are encoded so that the distortion of the synthesized speech is reduced, and these are transmitted together with the synthesis filter information.
An example is shown in which the present invention is applied to a voice encoding device by a voice encoding method represented by the CELP method.

In this embodiment, the present invention is applied in searching for a noise index, as will be described later. Before specifically explaining the configuration of FIG. 1, an index search procedure such as a noise index according to the present invention will be described with reference to FIG.

In setting the target vector (step S0), first, the number of index candidates is reduced based on the distortion evaluation of the code vector when there is no code error (step S1). That is, a smaller number of index candidates are selected from a large number of index candidates that can be the target vector. Next, the number of index candidates is further narrowed down based on the distortion evaluation that considers the quality deterioration due to the code error with respect to the reduced number of index candidates (step S2). In this way, one index for finally expressing the target vector is determined (step S3).

Next, the configuration of FIG. 1 will be described. The speech coding apparatus according to the present embodiment stores a past drive signal and generates a code vector from a designated pitch period, an adaptive codebook 2000, a noise codebook 2180 that stores a code vector corresponding to a noise index, and these. Of the synthesis filter 2 by applying a gain to each of the code vectors obtained from the codebooks 2000 and 2180 of
Gain circuits 2160 and 22 for generating drive signals for 270
50 and an adder 2260, a synthesizing filter 2270 that obtains a synthetic voice by inputting a driving signal, a synthesizing filter information analysis unit 2290 that analyzes the filter information of the synthesizing filter 2270 from the input voice, and a target vector from the input voice and the synthesizing filter information. Target vector creation unit 2300,
Using the target vector and the synthesis filter information, the distortion evaluation unit 2310 that obtains a distortion evaluation value for the synthetic speech obtained by the synthesis filter 2270, the code candidates (index candidates) are given to the codebooks 2000 and 2180, and the distortion evaluation unit 231.
A code selecting unit 2320 that selects a code candidate (index candidate) based on the distortion evaluation value obtained by 0, and a noise that considers a code error by narrowing down a small number of index candidates by the code selecting unit 2320. The code selecting unit 2325 for selecting an index and the code error adding unit 2326 for simulating a transmission code error and adding the index candidate from the code selecting unit 2325 are provided.

Hereinafter, each component in FIG. 1 will be described in more detail. Synthesis filter information analysis unit 2290
Analyzes the input voice, extracts and encodes the information of the synthesis filter that represents the outer shape of the spectrum of the voice, outputs this as synthesis filter information, and gives the parameters of the synthesis filter to the synthesis filter 2270. As an analysis method of the synthesis filter information, for example, LPC (Linear Predic
tion Coding) analysis method can be used.

Adaptive codebook 2000 is used to select the pitch period. That is, adaptive codebook 2000 stores past drive signals and selects a pitch cycle to be used as an encoding parameter from preset pitch cycle candidates. Specifically, the distortion between the combined vector candidate obtained by combining the code vectors obtained by giving the pitch cycle candidates to the adaptive codebook 2000 by the combining filter 2270 and the target vector generated by the target vector creation unit 2300 is The code selection unit 2320 selects the optimum pitch period on the basis of the evaluation that the value becomes smaller.

As a distortion calculation method, there is a method of actually calculating the distance distortion between the combined vector candidate and the target vector, but a value corresponding to the inner product of the combined vector and the target vector by modifying the distortion calculation formula, By combining the values corresponding to the power of the combined vector, it is possible to avoid calculating duplicate values that are fixed values for any pitch period candidate, and to reduce the amount of distortion with a smaller amount of calculation. The optimum pitch period can be selected by using the method of determining

For searching the adaptive codebook 2000, a codebook search method equivalent to setting the gain of the gain circuit 2160 to the optimum gain used in the normal CELP method can be used. At this time, the pitch period is searched for, assuming that the influence of the drive signal by the code vector extracted from the noise codebook 2180 is zero. If the number of candidates for the noise index can be limited to a very small number, a pitch period search is performed in consideration of the influence of the drive signal due to the code vector from the noise codebook 2180 to generate synthesized speech with less distortion. The effect that it is possible to search for possible pitch periods and noise codes can also be expected.

Next, the procedure for searching the noise index will be described with reference to the flowchart shown in FIG. The vector quantization according to the present invention is applied to the noise index search described here. In this case, the distortion evaluation unit 2310
Is used as first evaluation means for evaluating the distortion of the code vector without considering the code error of the noise index, and as second distortion evaluation means for evaluating the distortion of the code vector with consideration of the code error of the noise index. .

That is, first, the distortion evaluation section 2310 obtains a distortion evaluation result when there is no code error in the noise index information, and the first code selection section 23 is based on this distortion evaluation result.
A large number of noise index candidates preset in 20 are narrowed down to a small number of noise index candidates. Next, the distortion evaluation unit 2310 obtains a distortion evaluation result considering quality deterioration due to a code error of the noise index information, and based on the distortion evaluation result, the small number of noise index candidates selected by the first code selection unit 2320. About the second code selection unit 232
In step 5, the number of noise index candidates is narrowed down, and a noise index used for expressing the drive signal is searched.

Specifically, the first code selecting section 2320 uses the search loop 2340 to perform the noise codebook 2180.
To the noise vector candidate, and the code vector corresponding to this noise index candidate is added to the synthesis filter 22.
A small number of noise index candidates are selected with an index that the distortion between the combined vector candidate obtained by combining at 70 and the target vector generated at the target vector creation unit 2300 is small. As a method of calculating the distortion at this time, there is a method of actually calculating the distance distortion between the vectors, but the value corresponding to the inner product of the composite vector and the target vector by the modification of the formula of the distortion calculation and the power of the composite vector By combining equivalent values or values similar to this, it is possible to avoid calculating duplicate values that are fixed for any noise index candidate, and to reduce distortion with a smaller amount of calculation. You can select a small noise index.

The second code selection section 2325 narrows down the small number of noise index candidates selected by the first code selection section 2320 to a smaller number of index candidates. In the present embodiment, the second code selection unit 2325 narrows down to only one noise index, and finally obtains noise index information to be transmitted. At this time, the distortion evaluation value used in the distortion evaluation unit 2310 can be calculated using the same distortion evaluation value as that used under the condition without code error. However, in order to further simplify the calculation, a synthesis filter is used. Instead, a method of selecting an index is also effective, such as using distortion of the direct shape of the code vector due to a code error.

The code error adding unit 2326 simulates a transmission code error for each of a small number of noise index candidates selected by the second code selecting unit 2325 and gives a code vector when the index is changed by the code error. The noise codebook 2820 is indexed so that the distortion of can also be evaluated. As a method of evaluating the distortion of the code vector in consideration of the bit error of the transmission line, there is a method of using the expected value E of the distortion shown below.

[0033]

[Equation 1]

Here, E (i) is an expected value of distortion when the transmission path code corresponding to the index i is transmitted, p (j |
i) is the probability that the index i becomes the index j due to a code error in the transmission path, and d (j) is the distortion evaluation value when there is no code error in the code vector corresponding to the index j or a simplified distortion evaluation value. Represent For example, if the code on the transmission path corresponding to the index i is represented by n bits, the probability that 1 bit of these is wrong is ε.
= P (j | i) (where i ≠ j), and the probability of not making an error with n bits is 1-nε = p (i | i).

Of the small number of noise index candidates selected by the second code selection unit 2325, the index i that gives E of the most preferable size is selected as a noise index candidate narrowed down to a smaller number. When the strain evaluation value is defined so that d represents the strain amount, E is the expected value of the strain amount, so it is desirable to select the index when E becomes the minimum.

In order to simplify the calculation of E, the probability p
When the value of (j | i) is less than or equal to the threshold value, p is quantized in advance to 0 so that the value of p in Expression (1) with 0 is not calculated, or the value of p is fixed point DSP (digital signal). A method of quantizing to a value of (1/2) to the nth power (n is a natural number) is also effective so that the calculation becomes easy when a processor is used. Here, p (j |
The value of i) is assumed to be given in advance. The definition of the expected value E of the distortion of the equation (1) is also described in the document 2 introduced above, but in the document 2, the expected value E is used only in the design stage of the codebook, and the normal coding is performed. Uses a distortion evaluation value that does not consider code errors in the transmission path in the index search.

On the other hand, the present invention greatly differs from the above-mentioned conventional method in that in the ordinary encoding, the evaluation method considering the code error of the transmission path is incorporated in the index search. In this way, it is possible to select the index by incorporating the quality of the reproduced signal actually decoded under the condition of the code error into the evaluation of the index, so that the sudden quality deterioration due to the code error occurs. Can be made as small as possible.

One method of obtaining the expected value of distortion E (i) in the configuration of the speech coding apparatus shown in FIG. 1 is that the code error adding unit 2326 is applied to the index i given from the second code selecting unit 2325. The index j having a probability p (j | i) that cannot be ignored is given to the noise codebook 2180 as an index candidate, and the distortion evaluation values obtained for each code vector corresponding to j are weighted with respective probabilities to obtain the sum total. Is to take.

As another method for obtaining E (i),
The distortion evaluation value d for each index is temporarily stored in the memory in the first code selection unit 2320, and d necessary for the small number of index candidates selected by the code selection unit 2320 is read from the memory. Code selection unit 2
If the configuration is such that the expected value of the distortion evaluation value is calculated at 325, the code error adding unit 2326 in FIG. 1 becomes unnecessary.

When the index search method as described above is used, the index finally selected on the encoding device side can provide a high quality sound quality of a certain level or more when there is no error, and even when there is an error. Since it has been confirmed that the sound quality deterioration is small, even if a code error actually occurs on the transmission path, the possibility that the quality deterioration suddenly occurs on the decoder side can be minimized.

Next, an embodiment of a speech decoding apparatus for generating coded speech from the code information generated by the above speech coding apparatus will be described. FIG. 3 is a block diagram showing an embodiment of a speech decoding apparatus according to the present invention. In the figure, synthesis filter information, pitch period information, noise index information, and gain information, which are parameters encoded by the speech encoding device shown in FIG. 1, are input, and synthetic speech is generated from these information. The configuration is shown.

First, a method of reproducing the drive signal will be described. In adaptive codebook 1000, an adaptive code vector is obtained from a past drive signal based on pitch period information transmitted from the speech coder. For this adaptive code vector, the gain for the adaptive code vector obtained from gain codebook 1290 based on the gain information transmitted from the speech coding apparatus is used, and gain circuit 1160 is used.
To create the first vector by multiplying with the appropriate gain.

On the other hand, the noise code vector is extracted from the noise codebook 1180 based on the noise index transmitted from the speech coding apparatus, and the gain for the noise code vector obtained from the gain codebook 1290 is used for this to obtain the gain circuit 1250. Multiply the appropriate gain with
Generates a vector of.

Then, the adder 1260 reproduces the addition of the first vector and the second vector as a drive signal. Finally, this drive signal is input to the synthesis filter 1270 configured based on the synthesis filter information transmitted from the speech coding apparatus to perform speech synthesis, and the obtained synthesized speech is obtained from the output terminal 1280.

(Second Embodiment) FIG. 4 shows a vector coding apparatus according to the present invention which has a mechanism for obtaining code error information on a transmission line, or information on code error on a transmission line. 2 shows a second embodiment applied to a speech coding apparatus used in a wireless communication system capable of providing the coding side. In FIG. 4, the same parts as those in FIG. 1 are designated by the same reference numerals, and only the differences from the first embodiment will be described.

In this embodiment, the code error information detector 232 is used.
A command for updating the value of the probability p (j | i) according to the state of the code error by detecting the presence or absence of the code error or the degree of the code error by obtaining information on the code error of the transmission line from the terminal 2328. Is given to the code selection unit 2325. This has the effect that the index can be selected using a more accurate expected value of the distortion according to the error rate of the transmission path. For example, when the information that there is no code error in the transmission line is obtained, the degree of consideration of the code error is lowered, or the index search is performed based on the distortion minimization without considering the code error, and the code error is detected. When a large amount of information is obtained, it is possible to easily implement switching to an index search in which the degree of considering a code error is increased.

In the search of the noise index, a codebook search method equivalent to setting the gain of the gain circuit 2250 to the optimum gain used in the known CELP method can be used.

In this embodiment, the gain information is encoded by using the gain code book 2280 which can specify a specific gain by the gain information and the search loop 2350. In searching the gain information, the search is performed so that the distortion between the synthetic voice and the input voice is reduced.

Also in the present embodiment, as in the first embodiment, the first code selecting section 2320 temporarily stores the distortion evaluation value d for each index in the memory, and the code selecting section 2320. If the d necessary for the small number of index candidates selected in step 2 is read from the memory and the expected value of the distortion evaluation value is calculated in the second code selection part 2325, the code error addition part 2326 becomes unnecessary. Also,
A speech decoding apparatus corresponding to the speech encoding apparatus according to this embodiment may have the configuration shown in FIG.

In the above-described embodiment, in the index selection process, the number of index candidates is first reduced by distortion evaluation when there is no code error, and then the quality degradation due to a code error is performed for the reduced number of candidates. We tried to further narrow down the index candidates by distortion evaluation in consideration of, but if the codebook size is small and the number of all index candidates is small from the beginning, the indexes are evaluated from the beginning by distortion evaluation considering quality deterioration due to code error. It is clear that the method of selection is also effective for realizing a coding device that is resistant to transmission code errors.

Further, in the above embodiment, in the index selection process, the number of index candidates is first reduced by distortion evaluation when there is no code error, and then the quality reduction due to a code error is applied to the reduced number of candidates. Although the index candidates are narrowed down to one index by the distortion evaluation in consideration, the latter is narrowed down to a plurality of candidates, and the obtained plural candidates are, for example, distortion in another section of the audio signal. One index may be finally selected based on other distortion evaluation results such as the evaluation result and the distortion evaluation result considering the gain.

Furthermore, in the above-mentioned embodiment, the example in which the vector quantization according to the present invention is applied to the index search of the noise codebook has been described, but the present invention is not limited to this, and basically the parameter to which the vector quantization can be applied is applied. The present invention can be applied to the coded part of.

[0053]

As described above, according to the present invention, it is possible to provide a vector quantizer capable of minimizing the sudden quality deterioration of a reproduced signal even when there is a code error in a transmission line. You can

[Brief description of drawings]

FIG. 1 is a block diagram of a speech coding apparatus according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing the principle of the vector quantizer according to the present invention.

FIG. 3 is a block diagram showing an embodiment of a speech decoding apparatus according to the present invention.

FIG. 4 is a block diagram of a speech coder according to a second embodiment of the present invention.

FIG. 5 is a block diagram of a conventional speech coder.

[Explanation of symbols]

 2000 ... Adaptive codebook 2180 ... Noise codebook 2160 ... Gain circuit 2250 ... Gain circuit 2260 ... Adder 2270 ... Speech synthesis filter 2280 ... Gain codebook 2290 ... Synthesis filter information analysis unit 2300 ... Target vector generation unit 2310 ... Distortion evaluation unit 2320 ... 1st code selection part 2325 ... 2nd code selection part 2326 ... Code error addition part 2327 ... Code error information detection part 2330 ... Pitch period search loop 2340 ... Noise code search loop 2350 ... Gain code search loop

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location H03M 7/30 B 9382-5K 13/00 8730-5K H04B 14/04 Z H04N 7/24

Claims (3)

[Claims] 1. A vector quantizer for expressing a target vector by using a code vector specified by an index, and a distortion evaluating means for evaluating a distortion of the code vector in consideration of a code error of the index, and the distortion evaluating means. A vector quantization device, comprising: a selection unit that selects at least one index from a plurality of index candidates that can be used as an index for expressing the target vector based on the evaluation result of the evaluation unit. 2. A vector quantizer for expressing a target vector using a code vector specified by an index, wherein first distortion evaluating means for evaluating the distortion of the code vector and code error of the index are taken into consideration. A second distortion evaluating means for evaluating the distortion of the code vector, and based on the evaluation result of the first distortion evaluating means, a small number of index candidates that can be used as an index for expressing the target vector are selected. The first selection means for selecting an index candidate; and the first selection means based on the evaluation result of the second distortion evaluation means.
And a second selecting unit that selects at least one index from the index candidates selected by the selecting unit. 3. A vector quantizer for expressing a target vector using a code vector specified by an index, wherein first distortion evaluating means for evaluating the distortion of the code vector and code error of the index are taken into consideration. A second distortion evaluating means for evaluating the distortion of the code vector, and based on the evaluation result of the first distortion evaluating means, a small number of index candidates that can be used as an index for expressing the target vector are selected. The first selection means for selecting an index candidate; and the first selection means based on the evaluation result of the second distortion evaluation means.
Second selecting means for selecting at least one index from the index candidates selected by the selecting means, detecting means for detecting information relating to a code error in a transmission path for transmitting the information of the index, and detecting by this detecting means The second based on the information provided
3. The vector quantizing device according to claim 2, further comprising means for adjusting a degree of considering a code error in the selecting means.