JPH03211600A - Vector quantization method - Google Patents

Abstract

PURPOSE:To shorten the time for quantization by classifying the representative vectors by a zero cross number to plural pieces of groups, determining the zero cross number of the input vector to be subjected to quantization and calculating distortions with the representative vector within the corresponding group. CONSTITUTION:The zero cross number n of the input vector is calculated and if this n is nj-1<n<nj, this input vector is determined as to have the shape approximate to the shape of the j-th representative vector. The distortion di of all the representative vectors belonging to this group is calculated. The representative vector of the number (i) at which the distortion di is smallest is determined. This vector is determined as the representative vector for the input vector X and the number (i) of the decided representative vector is made the quantization value of the input vector X. Since the number of the representa tive vectors to be subjected to the calculation of the distortions is limited in such a manner, the time for the calculation processing at the time of determin ing the definite representative vector is shortened.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a vector quantization method for vector quantizing audio signals and the like.

2. Description of the Related Art In general, vector quantization is performed to quantize digital signals obtained by A/D conversion of audio signals and the like by computer processing.

FIG. 3 shows a flowchart of such conventional vector quantization processing. First, for the input vector X expressed as the following equation (1), the following (2
) M representative vectors Yi (where,
i = 1.2.3...M) is calculated by the following equation (3) for all i (step 1.2.3
).

X= (Xt, X2.X5−-・・,X5)−(1
)Vi= (Yll, Yiz, Yis..., Yi
N)...(2) Then, find the representative vector of number i that gives the smallest distortion di (step 4), decide it as the representative vector for the input vector X, and use step 5 (
In order to distinguish the representative vector determined in this way from other representative vectors, it is hereinafter referred to as a determined representative vector. ),
The number i of the determined representative vector is used as the quantization value of the input vector χ.

Figure 4 shows the relationship between the input vector x1 and the representative vector Yi when the number of components of the input vector is N=2 and the number of representative vectors is M=4, where 6 is the representative vector, 7 is the input vector, and 8 is the representative vector. is the boundary line between representative vectors.

To set the representative vector Vi, first use a training vector (hereinafter referred to as training vector) that has a distribution similar to the distribution of the large input vector, and apply it to the area of the required representative vectors YiO number M. Divide and determine the representative vector that minimizes the overall distortion.

In this way, input vectors can also be quantized using the conventional vector quantization method described above.

Problems to be Solved by the Invention However, in the above conventional vector quantization method, 1
Distortion must be calculated for all representative vectors each time quantization is performed, which increases the amount of calculation and makes real-time processing difficult.

The present invention solves these conventional problems, and aims to provide an excellent vector quantization method that can shorten the quantization processing time.

Means for Solving the Problems The present invention achieves the above objects by classifying representative vectors into a plurality of groups based on the number of zero crossings of the representative vectors and quantizing them when vector quantizing an audio signal or the like. By calculating the number of zero crossings of the input vector that performs This is designed to shorten the calculation processing time when determining a vector.

Therefore, according to the present invention, since the number of representative vectors to be subjected to quantization calculation is reduced and quantized, calculation processing time is shortened, quantization processing becomes faster, and real-time processing is possible. .

Embodiment FIG. 1 is a flowchart showing the procedure of a quantization method according to an embodiment of the present invention, and FIG. 2 is a diagram showing an example of group classification of representative vectors. In this embodiment, the number of zero crossings Qk of each vector is determined in advance for all representative vectors. In the case of a digital signal, since the amplitude of the signal does not always actually take a value of zero, one zero crossing is defined as when adjacent components have different positive and negative values.

Based on the number of zero crossings, the representative vectors are classified into L groups as shown in FIG. Then, the number of zero crossings is n
Let the representative vectors j−++l to nJ be the j-th representative group.

Next, the procedure for performing quantization will be explained.

First, the number n of zero crossings of the input vector is calculated (step 11). If this n is nj-tan≦nJ,
This input vector is determined to be close in shape to the j-th representative vector group (step 12), and distortion di is calculated only for all representative vectors belonging to this group (steps 13.14.15). The above equation (3) is used to calculate the strain di. Then, find the representative vector of number i that gives the smallest distortion di (step 16).
This is determined to be the representative vector for the input vector X (step 17), and the number i of the determined representative vector is set as the quantized value of the input vector X.

As described above, according to the above embodiment, if the number of representative vectors belonging to the j-th group is kj, the number of times to calculate the representative button, representative vector, distance, and distortion in this method is the first grouped L. This number of times is less than the number of calculations when comparing all representative vectors, and has the advantage that the calculation processing time when determining the final representative vector can be shortened.

Effects of the Invention As is clear from the above embodiments, when vector quantizing an audio signal or the like, the present invention classifies representative vectors into a plurality of groups based on the number of zero crossings of the representative vector, and also divides the input into which quantization is performed. By calculating distortion only for representative vectors within a vector, the number of representative vectors for which distortion is calculated is limited, so the calculation processing time when determining a final representative vector can be reduced. has.

[Brief explanation of drawings]

Fig. 1 is a flowchart showing a vector quantization procedure in an embodiment of the present invention, Fig. 2 is a diagram showing an example of classified representative vectors in the same embodiment, and Fig. 3 shows a conventional vector quantization procedure. The flowchart, FIG. 4, is a diagram showing an example of the arrangement of representative vectors in a conventional example.

Claims (1)

[Claims] When vector quantizing an audio signal, etc., the representative vector is classified into multiple groups based on the number of zero crossings of the representative vector, and the number of zero crossings of the input vector to be quantized is determined, and the number of zero crossings determined above is calculated. A vector quantization method characterized by calculating distortion only for representative vectors in corresponding groups.