JPH06138899A - Vector quantizer - Google Patents

Abstract

PURPOSE:To perform quantization with high accuracy with a small number of bits. CONSTITUTION:An extreme rear-end vector (N-th vector) is quantized by an extreme rear-end vector quantizing means 11. After that, each vector except the extreme rear-end vector of a block is predicted by performing interpolation by an interpolation differential vector generating means 13 by using both the quantization extreme rear-end vector of a preceding block obtained by a block unit delay means 12 and the quantization extreme rear-end vector of a current block, and a differential vector can be generated, and they are outputted by quantizing by an interpolation differential vector quantizing means 14, respectively. Decoding is performed at a decoder side by performing processing opposite to the above processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantizer which divides a series of vectors into a plurality of blocks and collectively quantizes the blocks.

[0002]

2. Description of the Related Art FIG. 2 shows the configuration of a conventional vector quantizer. In FIG. 2, reference numeral 21 is a vector quantizing means for quantizing an input vector on the encoding side,
Reference numeral 22 is a vector decoding means for decoding a vector from a vector quantization code on the decoding side.

The operation of the vector quantizer constructed as described above will be described below. For the input vector on the encoding side, the vector having the minimum distortion is selected from the table of vectors stored in advance by the vector quantization means 21 and the code (vector quantization code) is output. Then, on the decoding side, the vector decoding means 22 receives the vector quantized code obtained on the coding side as an input and decodes the vector corresponding thereto.

[0004]

However, in the above-described conventional vector quantizer, since each vector of the vector series is independently quantized, if the vector quantization is performed with a small number of bits, the quantizing precision is lowered. Had a problem.

The present invention is to solve the above-mentioned conventional problems, in which a vector series is grouped in a block unit composed of a plurality of vectors and the correlation between the vectors is used to quantize with a small number of bits and a high quantization accuracy. It is an object of the present invention to provide an excellent vector quantization device capable of performing the quantization.

[0006]

In order to achieve the above object, the present invention provides a tail vector quantizing means for quantizing the tail vector of each block of a vector sequence, and a tail vector quantizing means of the immediately preceding block. A block unit delay means for obtaining a vector in the current block, and each vector excluding the last vector of the block is predicted by interpolation from the last vector of the current block and the last vector of the previous block, and an interpolation difference for generating the difference vector A vector generation means and an interpolation difference vector quantization means for quantizing each interpolation difference vector obtained from this interpolation difference vector generation means are provided, and the decoding side decodes the last vector from the last vector quantization code. Tail vector decoding means and the interpolated difference vector quantization code of each vector excluding the tail vector. An interpolating difference vector decoding means for toll decoding, a block unit delay means for delaying the last vector by one block, a current block last decoding vector, a previous block last decoding vector, and each vector excluding the last vector from the interpolation difference vector. And a vector decoding means for decoding.

[0007]

According to the present invention, according to the above configuration, the vector series is grouped into a block unit composed of a plurality of vectors, the tail vector of each block is quantized, and the other vectors are tailed to the tail vector of the previous block and the current block. It is possible to utilize the correlation between the vectors by predicting by interpolation by using, and quantizing the difference vector, and it is possible to perform highly accurate quantization with a smaller number of bits. A vector series often has correlation between vectors, such as a parameter vector representing a vocal tract spectrum of speech, and in that case, it is less likely to be predicted by interpolation from preceding and following vectors as described above. High-precision quantization can be performed with the number of bits.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a vector quantization device according to an embodiment of the present invention. In FIG. 1, reference numerals 11 to 14 represent means on the encoding side, and 11
Is a tail vector quantization means for quantizing the tail vector of each block of the vector sequence, 12 is a block unit delay means for obtaining the tail vector of the previous block in the current block, and 13 is a tail end of the block. A plurality of interpolating difference vector generating means for predicting each vector excluding the vector by interpolation from the last vector of the current block and the last vector of the previous block and generating the difference vector thereof, 1
Reference numeral 4 denotes a plurality of interpolation difference vector quantizing means for quantizing the interpolation difference vector obtained from each interpolation difference vector generating means 13. On the other hand, 15 to 18 indicate means on the decoding side, 15 is the last vector decoding means for decoding the last vector from the last vector quantization code, and 16 is the interpolation difference vector quantum of each vector excluding the last vector. A plurality of interpolating difference vector decoding means for decoding the interpolating difference vector from the coded code, 17 a block unit delay means for delaying the last vector by one block, 18 a current block last decoding vector, previous block last decoding vector and interpolation difference vector Is a plurality of vector decoding means for decoding each vector except the last vector.

The operation of the vector quantizer configured as described above will be described below by taking as an example the case where N vectors are collectively quantized in block units. First, on the encoding side, the last vector (Nth vector) is quantized by the last vector quantization means 11. Then, using the quantized tail vector of the previous block and the quantized tail vector of the current block obtained by the block unit delay means 12, the interpolation difference vector generation means 13 interpolates each vector excluding the tail vector of the block. And the difference vector is generated.
Assuming that N = 2 (2 vectors x ₁ and x ₂ ), this difference vector d ₁ = {d (i), (i = 1, 2, ... M)} (M is the vector dimension number) is represented by an expression. And it becomes as follows.

[0010]

[Equation 1]

Next, on the decoding side, the tail vector decoding means 15 decodes the tail vector from the tail vector quantization code. Further, the interpolation difference vector decoding means 16 decodes the interpolation difference vector from the interpolation difference vector quantization code of each vector except the tail vector. Then, the vector decoding means 18 decodes each vector excluding the tail vector by using the previous block tail decoding vector, the current block tail code vector, and the interpolation difference vector obtained by the block unit delay means 17 as inputs.

As described above, according to the present embodiment, the vector series is grouped into a block unit composed of a plurality of vectors, the tail vector of each block is quantized, and the other vectors are quantized. It is possible to utilize the correlation between the vectors by predicting by interpolation using the last vector of the vector and quantizing the difference vector, and it is possible to perform highly accurate quantization with a smaller number of bits.

In the interpolation difference vector generation means 13 of the above embodiment, when performing prediction by vector interpolation, instead of using a constant value that does not depend on the order of the vector as the interpolation coefficient a, as in equation (1), , Interpolation coefficient vector a = {a (i), (i = 1,2, ... M)), which is the optimum vector from the table stored in advance.

[0014]

[Equation 2]To generate the difference vector by interpolation.
You may comprise so that it may be comprised. In this case, N = 2 (2
Cutle x₁, X₂) As the difference vector d ₁=
The expression {d (i), (i = 1,2, ... M)} is expressed as follows.
It

[0015]

[Equation 3]

[0016]

As described above, according to the present invention, a vector sequence is grouped into a block unit consisting of a plurality of vectors, the tail vector of each block is quantized, and the other vectors are divided into the previous block and the previous block. Prediction by interpolation using the last vector of the current block, and by quantizing the difference vector, the correlation between the vectors can be used, and it is possible to perform highly accurate quantization with a smaller number of bits. it can.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of a vector quantization device according to an embodiment of the present invention.

FIG. 2 is a schematic block diagram of a vector quantization device in a conventional example.

[Explanation of symbols]

 11 Last Vector Quantization Means 12 Block Unit Delay Means 13 Interpolation Difference Vector Generation Means 14 Interpolation Difference Vector Quantization Means 15 Last Vector Decoding Means 16 Interpolation Difference Vector Decoding Means 17 Block Unit Delay Means 18 Vector Decoding Means 21 Vector Quantization Means 22 Vector Decoding Means

Claims (2)

[Claims] 1. A vector sequence is divided into a plurality of groups,
On the coding side of the vector quantizer that collectively quantizes in block units, the tail vector quantizing means for quantizing the tail vector of each block of the vector series and the tail vector of the previous block are Block unit delay means for obtaining in the current block, and each vector excluding the tail vector of the block is predicted by interpolation from the tail vector of the current block and the tail vector of the previous block, and an interpolation difference vector generation for generating the difference vector Means, and an interpolation difference vector quantization means for quantizing each interpolation difference vector obtained from the interpolation difference vector generation means, the decoding side, the last to decode the last vector from the last vector quantization code Tail vector decoding means,
Interpolation difference vector decoding means for decoding the interpolation difference vector from the interpolated difference vector quantization code of each vector excluding the tail vector, block unit delay means for delaying the tail vector by one block, current block tail decoding vector, previous block And a vector decoding means for decoding each vector except the last vector from the last decoded vector and the interpolated difference vector. 2. An interpolating difference vector generating means on the encoding side selects an optimum vector from a table stored in advance as an interpolation coefficient vector when interpolating each vector, and performs interpolation according to the selected vector to perform decoding. The vector quantizing device according to claim 1, wherein the vector decoding is performed by performing similar interpolation using the coefficient vector selected on the encoding side.