JP2002533962A - Method and apparatus for detecting the optimal playback point - Google Patents

Abstract

(57) [Summary] In order to reduce the complexity of playback point detection in vector quantization, a group of playback points is selected near the predicted playback point, and distortion is applied to all of the selected playback points. The measurement is performed, and the point with the smallest distortion is selected as the reproduction point. Because the number of playback points to be evaluated is small, the complexity of the search is significantly reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

【Technical field】

The present invention relates generally to vector quantization, and more particularly, to a method and apparatus for detecting an optimal reproduction point from an N-dimensional point to be quantized.

[0002]

BACKGROUND OF THE INVENTION

When quantizing data, vector quantization is often performed. This is due to the fact that the deformation due to vector quantization is always smaller than the deformation due to scalar quantization. Furthermore, if the data to be quantized have a correlation, the deformation will be even smaller.

[0003] Vector quantization is complicated as an operation because it is necessary to detect an optimal vector. The optimal vector is usually defined as the vector with the smallest residual error. The complexity of the detection can be reduced by a list of "last values" or a similar method for detecting the reproduction point.

[0004] One field of technology that uses vector quantization is speech coding, where multiple correlated gain parameters are quantized simultaneously. All current CELP (Code Excited Linear Prediction) encoders try to minimize the mean square error in the weighted speech domain when quantizing many parameters. For best speech quality, the correlated gain parameters need to be quantized with respect to the distortion in the weighted speech domain, rather than the traditional distortion in the gain domain. This makes it possible to create a list of “most recent values”, since this list containing the distortion measurements depends on the signal, ie on values other than those to be quantized.

When the distortion measurement value depends on the signal, it is necessary to search all reproduction points in order to detect an optimum reproduction point. Digital American Mobile Phone System (D-AMPS)
The gain quantization of the new standard speech codec IS-641 used in E.C. is an example of this.

[0006] If all playback points have to be searched, the search is complicated. In the case of the CELP encoder described above, the distortion calculation itself is complicated, so that the complexity is particularly increased.

[0007]

[Summary of the Invention]

It is an object of the present invention to reduce the complexity of a reproduction point detection method and apparatus in vector quantization.

[0008] The object is to select a group of reproduction points near the expected reproduction point, evaluate the distortion for all of the selected group of reproduction points, and select from among the selected group of reproduction points. This is achieved by selecting the one with the least distortion as the playback point.

In this method, only a small number of reproduction points need to be searched.
Complexity is significantly reduced.

The present invention will be described in more detail below with reference to the accompanying drawings.

FIG. 1 exemplifies reproduction points of conversion gain parameters obtained from a speech codec. Conversion is performed to reduce the variance of the gain. If the variance is small, a quantizer with smaller distortion can be obtained. As is clear from FIG. 1, the two different gains have a strong correlation.

According to the present invention, the reproduction points shown in FIG. 1 are rearranged in advance in the N-dimensional space according to the distance from a reference point. FIG. 1 assumes that the reference point is in the upper right corner and N = 2.

Since the reproduction points of correlated data are gathered in a linear or curved shape in the N-dimensional space, a reference list is created so that a group of reproduction points surrounding any expected reproduction point can be accessed. Reference lists can be created. When the reproduction points are gathered in a straight line in the N-dimensional space as shown in FIG. 1, a reference list can be created by rearranging the reproduction points according to the distance from one end of the straight line.

FIG. 2 shows a conversion function for predictive value conversion for converting “distance from reference point” to “rearranged reproduction point” according to the case of this example. The conversion function shown in FIG. 2 that is created in advance is used to predict the reproduction point that calls the “rearranged VQ index” in FIG. 2 from the “distance from the reference point” as described below. I do.

As described below, a conversion function is created for each of the rearranged reproduction points.

The conversion function shown in FIG. 2 is reported by two or three linear linear equations that define an uncomplicated conversion function that converts “distance from reference point” to “reordered playback point”. Can be.

According to the present invention, detection of an N-dimensional optimum reproduction point to be quantized corresponding to the N parameters extracted from the input signal is performed in the following manner.

Using the non-quantized value of the N-dimensional point to be quantized, the distance from the reference point is calculated according to the following equation.

(Equation 1) Here, N the number of dimensions of space, X _i is quantized not the value of i dimension, Ref _i is the reference point of the i-dimensional.

Next, using the linearized version of the conversion function shown in FIG. 2, the calculated distance is converted into “rearranged reproduction points”.

Then, a group of reproduction points surrounding the predicted reproduction point is selected from the rearranged reproduction points.

The number of selected playback points must always be large enough to include the optimum playback point therein.

This is achieved by performing a search for the optimum reproduction point by the conventional method, that is, for all reproduction points and by the method of the present invention, and confirming that the same optimum reproduction point is always found by both methods. It can be carried out.

In the case of the example shown in FIG. 1, in order to ensure that the optimum reproduction point is always included in the selected group of reproduction points, the entire reproduction point is placed on both sides of the expected reproduction point. 10
It is sufficient to select a playback point of%.

As described above, the number of reproduction points to be selected can be determined in advance.

For each reproduction point included in the selected group of reproduction points, a so-called distortion value is calculated.

The distortion values represent the original signal, the speech signal coming into the speech coder, and the reproduced signal, the output signal from the speech coder.

If the value of the distortion is more dependent on one dimension than on the other, this point can be taken into account by weighting each dimension in the above-described distance calculation.

Finally, from among a group of selected reproduction points around the predicted reproduction point, a reproduction point at which the calculated distortion value is minimized is selected as an optimum reproduction point.

In the above example, only a very small portion of the total number of reproduction points needs to be searched, so that the complexity of the search is reduced by about 80%.

The complexity is further reduced if it is permissible that it is not always necessary to use the optimal reproduction point. Complexity can be further reduced by reducing the number of selected playback points.

The cost of determining the expected value is only a small amount of memory needed to store the reference value used to access the reordered playback points according to distance. When playback points need to be transmitted with bit errors over the communication channel, the playback points are often reordered to minimize the distortion caused by one bit error. This method is called index assignment. If it is not necessary to assign an index to the reproduction point, the reproduction points can be arranged in this way from the beginning, so that a reference list is unnecessary.

In the example shown in FIG. 3, reproduction points are substantially uniformly collected in a two-dimensional rectangle. In this example, there is no correlation between the dimensions. Let us reduce the order of the reference points so that we can find the correlation. In this case,
The space of the first dimension. In FIG. 3, the reference point is provided at the position of the mark x on the right side, and only the horizontal plane distance is used. However, also in this case a significant reduction in complexity is obtained.

In FIG. 4, the reproduction points are generally gathered on the circumference. In this case, as the “distance”, a distance related to an angle, that is, an angle formed with an expected reproduction point can be used. Assuming that the expected playback point is in this case 0 or 2π radians, the selected set of playback points must include points near 0 and 2π radians. Also in this case, the complexity of the search can be reduced by determining the reproduction point in this way.

As is apparent from the above description, the present invention reduces the complexity of detecting the optimum playback point. It is obvious for a person skilled in the art that the method described above can also be performed by a processor.

[Brief description of the drawings]

FIG. 1 illustrates a reproduction point of a two-dimensional vector quantizer in which values of two dimensions have strong correlation.

FIG. 2 shows a relationship between an expected value and a rearranged reproduction point for the reproduction point shown in FIG. 1;

FIG. 3 illustrates a reproduction point of a two-dimensional vector quantizer in which values of two dimensions are distributed on average.

FIG. 4 illustrates another example of a reproduction point of a two-dimensional vector quantizer where values of two dimensions are gathered along a boundary of a circle.

──────────────────────────────────────────────────続 き Continuation of front page (81) Designated country EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, IT, LU, MC, NL, PT, SE ), OA (BF, BJ, CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, GM, KE, LS, MW, SD, SL, SZ, TZ, UG, ZW), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AL, AM, AT, AU, AZ, BA, BB, BG, BR, BY, CA, CH, CN, CR, CU, CZ, DE, DK, DM, EE, ES, FI, GB, GD, GE, GH, GM, HR, HU, ID , IL, IN, IS, JP, KE, KG, KP, KR, KZ, LC, LK, LR, LS, LT, LU, LV, MA, MD, MG, MK, MN, MW, MX, NO, NZ, PL, PT, RO, RU, SD, SE, SG, SI, SK, SL, TJ, TM, TR, TT, TZ, UA, UG, UZ, VN, YU, ZA, ZW

Claims (4)

[Claims] 1. An N-dimensional space containing a predetermined number of reproduction points, used by a decoder to decode an input signal encoded by the encoder, where N ≧ 2, extracted from the input signal. A method for detecting an optimal reproduction point for an N-dimensional point to be quantized, corresponding to N parameters, comprising:-rearranging a predetermined number of reproduction points in advance according to a distance from a reference point; Creating a function that converts the distance from the reference point to a reordered playback point; calculating the distance from the reference point to the N-dimensional point; and calculating the distance calculated by the created function, Convert to a predicted playback point whose distance from the reference point is closest to the distance to be converted;-select a sorted group of playback points around the expected playback point;-all of the selected group of playback points. Play Calculating the distortion value of the point; selecting the reproduction point with the smallest calculated deformation value as the optimal reproduction point in the selected group of reproduction points. 2. The method according to claim 1, wherein a pre-defined set of reordered reproduction points around the expected reproduction point is selected. 3. An N-dimensional space containing a predetermined number of reproduction points, N ≧ 2, used by a decoder to decode an input signal encoded by the encoder, the N signal being extracted from the input signal. An apparatus for detecting an optimal reproduction point for an N-dimensional point to be quantized corresponding to N parameters, comprising: means for rearranging a predetermined reproduction point in advance according to a distance from a reference point; Means for creating a function for converting the distance from the reference point to the reconstructed reproduction point in advance;-means for calculating the distance from the reference point to the N-dimensional point;-calculation by the created function Means for converting the obtained distance to a predicted reproduction point whose distance from the reference point is closest to the distance to be converted;-means for selecting a rearranged group of reproduction points around the predicted reproduction point; Calculating the distortion values of all the reproduction points of the selected group of reproduction points; selecting the reproduction point having the smallest calculated deformation value as the optimum reproduction point in the selected group of reproduction points; An apparatus comprising means for performing 4. The apparatus according to claim 3, wherein said selecting means selects a predetermined number of rearranged reproduction points around an expected reproduction point.