JP3049574B2 - Gain shape vector quantization - Google Patents

Abstract

PURPOSE:To efficiently encode the signal series of voices and images to a mode of digital with a small amount of information while suppressing the distortion caused by the encoding. CONSTITUTION:At the time of retrieving the shape vector code note in the decision procedure of the shape vector, all the possible combinations of a pair of shape vectors are retrieved with a smaller amount of operation by using the ratio of the gain of respective shape vectors assumed in advance. For example, in the retrieval of shape vector code notes 21 to 23, each combination of the shape vector is multiplied by the retrieval gain indicated by a shape vector retrieval gain vector 11 to obtain sum of vectors. The sum of vectors is multiplied by the optimum gain so as to minimize the distortion between the sum of vectors and the vector to be encoded, then a pair of vectors to make the distortion minimum is selected.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for digitally encoding a signal sequence of voice or image, and more particularly to a method for efficiently performing digital encoding with a small amount of information while suppressing distortion due to encoding. The present invention relates to a gain shape vector quantization method.

[0002]

2. Description of the Related Art As a method for efficiently encoding a signal sequence such as a voice or an image, vector quantization using a vector obtained by batching a plurality of samples as a quantization unit is often used. In this method, first, N samples are collectively regarded as an N-dimensional vector, and this is used as a unit of quantization. Then, a codebook including a large number of quantized representative values of N-dimensional vectors is prepared, and a vector in the codebook having the minimum distance from the vector to be quantized is set as the quantized value. This codebook is usually created from a large amount of learning data by the Lloyd method, the LBG method, or the like. If this vector quantization method is used, it is possible to encode a signal sequence at a low bit rate and with relatively small distortion. Here, in order to keep the quantization distortion small for various types of signal sequence vectors to be expressed, the codebook size may be increased. However, when it exceeds a certain level, the amount of calculation and the amount of codebook memory in search become enormous, and a large amount of data for codebook learning is required, which is not practical. In such a case, the information on the vector to be coded is divided into a shape vector representing the shape and a gain that multiplies the entire shape vector by a fixed factor, and a gain shape (Gain-Shape) expressing the reproduction vector with codes representing the two. Vector quantization may be useful. Furthermore, in order to make the amount of codebook memory and the amount of data for codebook learning a feasible level, the shape vector codebook is divided into a plurality, and the shape vector codebook output and the gain corresponding to each output are used. A method of expressing a reproduction vector by weighted linear combination is considered. The quantization value (code) determination method includes the following (1) and (2). (1) The gain of the number of dimensions corresponding to the number of shape vector codebooks is regarded as a vector (hereinafter, referred to as a gain vector), and the value is quantized on a vector basis to prepare a codebook. Then, as shown in FIG.
In the gain codebooks 21 to 24, the shape vectors (plural)
And all the combinations of the gain vectors are determined, and the combination that minimizes the distortion is determined. FIG. 2 shows a case where there are three shape vector codebooks. (2) When searching for a shape vector codebook, the order is determined for a plurality of codebooks, and the codes are searched one by one in order from the top to determine the optimal code. At this time, the optimal gain logically obtained from the shape vector and the vector to be coded, that is, the gain that minimizes distortion is used. In addition,
When searching for a single shape vector codebook, the influence of the shape vector determined before that is removed by subtracting it from the vector to be coded. Further, gain encoding can be performed by a method of performing scalar quantization on a theoretically optimum value obtained at the time of shape vector search, or a method of preparing a gain vector codebook and performing the search. FIG. 3 shows an example of a shape vector codebook search by this method (2). Here, three shape vector codebooks (A to C)
A case where a second shape vector codebook B22 of 21 to 23 is searched will be described. In this case, the shape vector codebook A21
Are determined before searching the shape vector codebook B22. The gain for minimizing the distortion is multiplied by the calculator 26. An example of the gain vector codebook search is shown in FIG. 4, in which the value of the gain vector codebook 24 corresponding to the output obtained by searching the shape vector codebooks 21 to 23 is obtained. Related methods of this type include, for example, Japanese Patent Application No. Hei.
-166831.

[0003]

In the above-mentioned method (1) in the prior art, since all combinations are searched, an optimum code is obtained. However, the number of combinations is usually enormous and impractical. Further, in the method (2), since the search is not performed by using all combinations of the shape vector sets, it is inevitable that the quantization distortion increases more than the code obtained by the optimal combination of the shape vectors. An object of the present invention is to improve such a problem and to provide a gain shape vector quantization method capable of quantizing a signal sequence having a small distortion with a small amount of calculation and a small memory capacity.

[0004]

To achieve the above object, a gain shape vector quantization method according to the present invention comprises a gain codebook and a plurality of shape vector codebooks, and a gain ratio between the shape vector codebooks. Tentatively fixed, take a weighted linear sum for each shape vector codebook accordingly, multiply the sum vector by a gain that minimizes the distortion between the obtained vector and the encoded vector, After selecting the shape vector set that minimizes the distortion and determining the shape vector set, the gain codebook is searched by the value of the weighted linear sum using the set to determine the optimal gain set. There is a feature in the decision.

[0005]

In the present invention, when searching for a shape vector codebook in the shape vector determination procedure, all possible combinations of shape vector sets are searched for by using the gain ratio of each shape vector assumed in advance. This makes it possible to greatly reduce the amount of calculation for the search as compared with the case where all the combinations of the shape vector gains are performed by the conventional method. Also, as compared with a method of sequentially searching and determining the codebook of the shape vector based on the previous result, and then searching the gain vector, quantization with less distortion is possible. Therefore, a code that is close to optimal can be found with a smaller amount of calculation.

[0006]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is an explanatory diagram of a shape vector codebook search in the first embodiment of the present invention. FIG. 5 is a flowchart showing a vector quantization process in the first embodiment of the present invention.
FIG. 4 is an explanatory diagram of a gain vector codebook search in the first embodiment of the present invention. The vector quantizer of the present embodiment includes a plurality of shape vector codebooks and gain vector codebooks, and an arithmetic unit for performing addition / multiplication, constant multiplication, and the like. In FIG. 1, reference numeral 11 denotes a gain vector fixedly set for shape vector search, 12 denotes an arithmetic unit for giving an optimum gain corresponding to the shape vector, 21 to 23 denote three shape vector codebooks A to C, 24 Is a gain vector codebook. In the present embodiment, three shape vector codebooks 21 to 2
When the optimum vector code is determined from 3, the gain ratio is fixed (for example, 1: 1: 1) by a three-dimensional shape vector search gain vector 11 as shown in FIGS.
(501), three shape vector codebooks 21 and 2
A search is made for all combinations of No. 3 (502). Specifically, each of the combinations of the vectors is multiplied by the search gain indicated by the shape vector search gain vector 11, and the sum of the vectors is calculated as the distortion between the sum vector and the encoded vector. Arithmetic unit 1 for optimal gain to be minimized
Multiply by 2 and select a set of shape vectors that minimizes the distortion. Note that the gain vector codebook 24 is not used in the shape vector codebook search. Further, when searching the gain vector codebook 24, as shown in FIGS. 5 and 6, the set of shape vectors is fixed as the optimal vector code (503), and the entire search of the gain vector codebook 24 is performed (504). , 505). This involves multiplying each of the determined shape vectors by the value of the corresponding dimension of the gain vector, taking the vector sum, and selecting the gain vector that minimizes the distortion between the vector and the encoded vector.
In the gain vector codebook search, the shape vector search gain vector 11 is not used. An optimum gain shape vector is determined from the gain vector thus determined and the shape vector (506). In this embodiment, an example in which the number of shape vector codebooks is three has been described, but it is apparent that the present invention can be applied to any case as long as there are a plurality of shape vector codebooks.

FIG. 7 is a block diagram showing a part of a speech coding apparatus according to a second embodiment of the present invention. 7, reference numeral 71 denotes a speech analysis unit, 72 denotes an adaptive codebook, and 73 and 74.
Are noise codebooks A and B, 75 is a gain vector codebook, 76
Reference numeral 78 denotes a linear prediction synthesis filter, 79 denotes a distance (distortion) calculation unit, 80 denotes a codebook search control unit, and performs speech encoding of a CELP (code excited linear prediction) method by such a configuration. In this embodiment, a case will be described in which the gain shape vector quantization method shown in the first embodiment is applied to the determination of the drive excitation vector of the linear prediction synthesis filter. In this case, first, the time series of the audio signal to be encoded is divided into units called frames corresponding to the vector to be encoded. Next, the speech analysis unit 71 performs linear prediction on each of the in-frame speeches (time-series vectors), calculates a linear prediction parameter representing the spectrum envelope shape of the speech, and performs linear prediction in frame units based on the parameters. The prediction synthesis filters 76 to 78 are configured. Here, the gain shape vector quantization method shown in the first embodiment is introduced for determining the excitation vector for driving the linear prediction synthesis filters 76 to 78. The codebook corresponding to the shape of the driving excitation vector in the present embodiment outputs, as in the normal case, a repetition of the time series of the driving excitation in the previous frame at a certain cycle (this cycle corresponds to the code). The adaptive codebook 72 represents a periodic component of speech, and the noise codebooks 73 and 74 represent noise components of speech. The outputs of these codebooks passed through linear prediction synthesis filters 76 to 78 are regarded as the shape vectors in the first embodiment, and the driving excitation vector codebooks 72 to 74 and the corresponding gain vector codebooks 75 are used. Perform a search. In this case, the shape vector codebook to be searched may be for only the noise codebooks 73 and 74, or may be for both codebooks 72 to 74. Note that the difference from the first embodiment is that the driving excitation vector is passed through different linear prediction synthesis filters 76 to 78 for each frame to be the reference shape vector, and that the fixed excitation is fixed in each quantization. In addition to the codebook prepared in advance, the adaptive codebook 72 which outputs a periodicized driving excitation time series of the previous frame as an output can be a search target of the shape vector codebook.

[0008]

According to the present invention, it is possible to greatly reduce the amount of calculation for retrieval as compared with the conventional method (1) for performing all combinations of shape vector gains. For example, when the shape vector codebook is composed of 3 bits (one codebook is composed of eight vectors) × 4, and the gain vector codebook is composed of eight bits (one codebook is composed of 256 vectors), The number of times that the vector is compared with the vector to be quantized is the following ratio in the conventional method (1): the conventional method (2): the method of the present invention.

(Equation 1) Further, as compared with the conventional method (2) of searching and determining the codebook of the shape vector in order based on the previous result and then searching the gain vector, the search can be performed by all combinations of the shape vectors, so that the distortion can be further reduced. Is possible. Therefore, a quantization result with less distortion can be obtained with a smaller amount of calculation.

[0009]

[Brief description of the drawings]

FIG. 1 is an explanatory diagram of a shape vector codebook search in a first embodiment of the present invention.

FIG. 2 is an explanatory diagram of a conventional shape gain vector quantization method.

FIG. 3 is an explanatory diagram of a conventional shape gain vector quantization method.

FIG. 4 is an explanatory diagram of a conventional shape gain vector quantization method.

FIG. 5 is a flowchart illustrating a vector quantization process according to the first embodiment of the present invention.

FIG. 6 is an explanatory diagram of a gain vector codebook search in the first embodiment of the present invention.

FIG. 7 is a configuration diagram illustrating a part of a speech encoding device according to a second embodiment of the present invention.

[Explanation of symbols]

 Reference Signs List 11 gain vector for shape vector search 12 arithmetic unit 21 shape vector codebook A 22 shape vector codebook B 23 shape vector codebook C 24 gain vector codebook 25 arithmetic unit 26 arithmetic unit 71 voice analysis unit 72 adaptive codebook 73 noise code Book A 74 Noise codebook B 75 Gain vector codebook 76 Linear prediction synthesis filter 77 Linear prediction synthesis filter 78 Linear prediction synthesis filter 79 Distance (distortion) calculation unit 80 Codebook search control unit

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-64-74883 (JP, A) JP-A-3-32228 (JP, A) JP-A-5-14208 (JP, A) (58) Field (Int.Cl. ⁷ , DB name) H03M 7/30

Claims (1)

(57) [Claims] 1. A system comprising a gain codebook and a plurality of shape vector codebooks, wherein a signal sequence obtained by dividing a signal sequence into a plurality of sample vectors is used as a quantization unit, and a quantized value of the vector is defined as a plurality of sets of shape vectors. In a gain shape vector quantization method expressed as a weighted linear sum with a set of gains corresponding to the shape vector, a gain ratio between the shape vector codebooks is fixed, and a shape vector codebook is formed according to the gain ratio. A weighted linear sum is calculated for each of the shapes, a search is made for all possible combinations of the set of shape vectors, an optimal vector code is determined, and then the above gain codebook is calculated using the value of the weighted linear sum using the vector code. A gain shape vector quantization method characterized by searching and determining a set of gains.