JP2002503834A - System and method for split vector quantized data encoding - Google Patents

Abstract

(57) A method and system for performing split vector quantization used in determining constrained ordered set values such as line spectrum pair parameters to determine spectral parameters in a data compression system comprises: Utilize multiple codebooks (22a-22c) that contain delta-encoded bounded ordered set values normalized to the upper and lower bounds. The LSP reconstructor (34) decodes data such as speech based on the normalized delta quantized data of the line spectrum pair parameters obtained from the split vector reconstruction codebook (22a to 22c). , The received spectral parameters are reconstructed. The LSP reconstructor (34) dynamically generates line spectrum pair parameters based on the normalized delta quantized data. In another embodiment, instead of storing the absolute values of the line spectrum pair parameters in a segmented codebook, the combination of at least two magnitude vectors and at least one normalized delta quantization vector is It is used for conversion.

Description

DETAILED DESCRIPTION OF THE INVENTION

The present invention relates generally to data encoding systems and methods that utilize scalar or vector quantization of a constrained ordered set, and more particularly to audio or video encoding systems. Order parameters such as line spectrum pair parameters to determine spectral or other parameters as used in
rs) is used to determine the set of split vector quantization (split vector
quantization).

BACKGROUND OF THE INVENTION Quantization of constrained ordered sets is commonly used in modern speech and speech compression systems to encode sets of ordered parameters, such as line spectrum pair parameters. The line spectrum pair parameter is a linear predictive coding (linea
(r predictive coding) coarse spectral information (coarse
spectral information). Appropriate stability constraints
constraints), the linear predictive coding analysis for the line spectrum pair parameter transformation generally produces a rank ordered set with good behavior, ranging from 0.0 to 0.5.
(rank ordered set). The correlation properties of line spectrum pairs make them attractive for the various scalar and vector quantization methods used in some recent speech compression standards.

For example, the American Telecommunications Industry Association (TIA) dated January 1997
IS-127 standard "Enhanced Variable Rate"
Codec, Speech Service Option 3 For Wideband Spread Spectrum Digital Syst
ems "utilizes a technique called" weighted split vector quantization "that is used to quantize LSP parameters. The 10-dimensional line spectrum pair vector is a multi-subve
ctors) or segments, which are the codebook memory perspective,
Easy to manage, both in terms of code search complexity and code search. For example, the IS-127 rate 1/2 codec employs a 22-bit, 3-segment weighted split vector quantization scheme. The 10-dimensional LSP vector is divided into three segments, resulting in 3 + 3 + 4 elements. This scheme utilizes a 7,7,8 bit designation for each vector segment codebook. A separate codebook is used for each segment. Therefore, this method requires a ^{3x2 7 + 3x2 7 + 4x2 8} = 1792 words of the codebook memory (ROM), each of which requires a weighted square error (weighted squared error) calculated. This method requires 42 megawords of storage and a brute force calculation that requires a weighted squared error calculation.
Although the amount of codebook memory is reduced by dividing the vector into codebook segments as compared to a force-based 22-bit 10-element code vector design, the weighted split vector quantization scheme does this brute force. Does not exhibit the same performance effects as the method. One reason is that there is overlap between vector segments, such that some codebook data combinations do not retain the LSP order property, and thus are not valid for a given linear predictive coding coefficient. It is not to be. The effect is that the code vectors in the codebook are substantially pruned, thereby reducing the vector quantization coding gain.

[0003] For example, FIG. 1 shows a sample distribution diagram of LSP parameters. Since the LSPs are given in a strictly ascending rank order set, the first element in the second code vector must be greater than the last element in the first code vector. 3-3-4 If the optimal code vector of the first segment of WSVQ is found to be {ω ₁ , ω ₂ , ω ₃ }: {0.10,0.15,0.20}, the second code vector the first element ω ₄ of the stomach, such must be greater than 0.20 (ω _4> 0.20). However, due to the statistics in the LSP training database and the independent training of the codebook for each segment, there may be too many vectors (entries) in the codebook that do not satisfy this condition. The effect is that code vectors in the codebook that do not satisfy this condition are invalidated, rejected as possible candidates by the encoder, and the codebook is substantially pruned.

[0004] A known method that attempts to address this problem employs a weighted split vector quantization scheme, where the codebook values are the current and previous quantized line spectral pairs. Based on the difference between In effect, the system constructs the current quantized line spectrum pair value by adding the specified codebook value (△ ω _i hat) to the previously constructed line spectrum pair:

[0005]

(Equation 1)

This method, known as delta decoding, generally removes codebook element duplication (as long as delta omega is strictly positive), but similarly if each delta value is relatively large, A severe pruning effect may occur.
For example, if the last quantized LSP value (ω _i-1 hat) is 0.3 and the value of the current LSP delta codebook (△ ω _i hat) extends to (0.05, 0.30) , 0.2 are invalid because the LSP value is greater than or equal to 0.5. Furthermore, codebook values approaching 0.2 are substantially "prune" because the probability of the corresponding LSP occurring in this space is greatly reduced. Again, this pruning may cause a degradation in the overall quantization performance in the speech coding system, which may occur at any stage of the process. There are also other methods that attempt to remove redundancy from line spectrum pair parameters using interframe prediction,
In general, these methods are more susceptible to inherent channel errors in existing wireless communication systems.

Another system of LPC coefficient quantization in speech coding, known as lattice quantization, is described in the 1995 International Conference on Acoustics, Speech and Signal Processing (1995).
Minjie Xie et al. in the IEEE bulletin from Peech and Signal Processing)
Fast and Low-Complexity LSF Quantization using Algebraic Vector Quantize
r "(1995). This technique utilizes a 4-segment split vector quantizer with a maximum segment length of 3 to provide 10 frames per frame.
LSPs are quantized. The components of one and only segment are absolute L
The remaining segments represent the LSP difference normalized to add to the value for which the sum of the components is not greater than 1 for each segment. One stored codebook is used to provide a codebook entry for one segment that represents an absolute LSP value, while the codebook entries for the remaining segments use a code vector index and an implicit grid structure (implicit grid structure). lattice
structure) in real time. Using a lattice structure
Less memory and search complexity is required compared to using only stored codebooks. Also, utilizing a normalized LSP difference in such a system can prevent pruning from occurring. However, while normalization helps ensure that all codebook entries are valid, normalization can realign data, resulting in property changes from the original parameters. Sometimes.

[0008] For example, utilizing the normalized delta quantization values affects the quantization error by changing the distribution of the normalized delta quantization values and other statistics, so that these statistics are This is different from the derived absolute value statistics. Normalized delta
In a codebook containing entries, the normalized delta LSP values are mapped to a fixed range. The mapping from absolute LSP values to the normalization range is non-linear because the values of the lower and upper normalization limits of a given LSP to be normalized are a function of previously quantized frame data. This non-linear mapping (warping) has the effect of altering the distribution of the normalized delta values and other statistics, so that these statistics are different from the statistics of the absolute values from which they were derived. These differences are
This may affect the design of the entry and the resulting quantization error. This warping effect may result in performance gains or losses depending on the statistics of the data under consideration. If the loss due to warping occurs, the gain due to pruning rejection must be greater than the loss due to warping in order to utilize the normalized delta value to achieve an overall reduction in quantization error. No. If the loss due to warping negates the gain due to pruning arrest, the use of absolute values can be prioritized. Therefore, judging the LSP parameter from the normalized vector may lower the coding performance.

Furthermore, the normalization applied in such a system is generally restricted so that the sum of the normalized components in each segment cannot be greater than one.
This restriction is necessary to enable the use of triangular and simplex lattice structures. However, restricting the encoding system to this lattice structure can wastefully limit encoding performance.

The use of a codebook entry based on the lattice structure of Xie et al. Can be suboptimal for quantization of LSP data. By restricting the codebook entries to points on a regular grid structure, the codebook entries do not optimally fit the statistical distribution of the data being quantized, thereby increasing the average quantization error. In contrast, utilizing a stored codebook can optimally fit a codebook entry to a set of training data having a statistical distribution that matches the statistical distribution of the data being quantized, Helps reduce the average quantization error. Thus, for a given number of codebook entries, the performance of the trained and stored codebook is generally superior to that of codebooks based on extremely complex lattice structures with stringent memory requirements.

[0011] Thus, helping to reduce processing time, helping to reduce pruning,
What is needed is a method and system for performing improved split vector quantization in a data compression or coding system that can improve coding performance in constrained ordered set quantization.

Description of the Preferred Embodiment To determine the spectral parameters in a data compression system, a split vector quantization scheme used in determining constrained ordered set values, such as line spectral pair parameters, is used. The method and system for doing this are in the constraint space (constraint sp
ace), a multiplexed codebook (mul) containing delta-encoded binding order set values, such as line spectrum pair values, normalized to upper and lower limits, such as the effective dynamic range of
Use tiple codebooks). The quantized value in the segmented codebook is the last quantized bounded set value, such as a line spectrum pair value, and another higher (or highest possible), such as a line spectrum pair value. Represents the percentage of the distance from the bound order set value. Therefore, all values stored in codebook memory will be valid LSP values, thus preventing codebook pruning. As used herein, split vector quantization includes scalar quantization where the vector length is one.

[0013] The LSP reconstructor accesses the normalized delta quantized data in the segmented codebook, which is used to decode speech or other data by using a line spectrum pair. Used to determine the bound order set value of a parameter, etc. The LSP reconstructor reconstructs the received spectral parameters based on the reconstructed code data, i.e., the normalized delta quantized data of the line spectrum pair parameters obtained from the split vector reconstruction codebook. To construct. The LSP reconstructor dynamically generates line spectrum pair parameters based on the normalized delta quantized data. Next, the reconstructed parameters are sent to a short term synthesis filter to decode data such as speech.

In a preferred embodiment, instead of storing the absolute values of line spectral pair parameters in a segmentation codebook, the disclosed system and method uses at least two magnitude vectors and spectral quantization. And combination of the normalized delta quantized data with at least one vector for use.

FIG. 2 shows an example of encoded input data representing encoded speech via a speech parameter 14 such as a spectral parameter to reconstruct speech transmitted on an air interface or channel. 1 illustrates a speech decoder 10 in a communication system, such as a wireless telephony communication system, for receiving a speech signal. Speech parameters 14 are decoded by a parameter decoder as is well known in the art. The speech decoder 10 includes, for example, a fixed codebook with associated vectors as known in the art, a long term prediction block 18 as known in the art, and a short term LPC synthesis filter. (short term prediction LPC synthesis filter) 20. The short-term prediction LPC synthesis filter 20 uses the split vector LSP reconstruction codebook 2
L generated based on LSP data derived by using 2a to 22c
Use PC information.

The split vector LSP reconstruction codebook 22 a-22 c is a split containing reconstruction code data representing normalized delta quantized data for determining line spectrum pair parameters or other constrained ordered set parameters. -Functions as a vector reconstruction code source. Each entry of the reconstructed code data in the codebooks 22a-22c corresponds to valid data for determining constrained ordered set data, such as line spectrum pair parameters, and receives received encoded input data 12 or other data. Facilitates reconstruction of spectral parameters. The speech parameters 14 include, among others, a fixed codebook index parameter 24, a fixed codebook gain parameter 26, a long-term delay parameter 28, a long-term prediction gain parameter 30, and a normalization delta quantization (NDQ).
a quantization) codebook index parameter 32.

The NDQ codebook index parameter 32 indexes a vector in one or more of the split vector LSP reconstruction codebooks 22a-22c. The line spectrum pair parameters are reconstructed by the LSP reconstructor 34 based on the indexed normalized delta quantized data in the split vector LSP reconstruction codebooks 22a-22c. LS
A P / LPC transformer (LSP to LPC transformer) 36 converts LSP parameters into LPC information, as is well known in the art. The LPC information is then used by the conventional short-term prediction LPC synthesis filter 20, along with the long-term prediction information, to generate output speech.

Specifically, the fixed codebook index parameter 24 is used to determine the stochastic component of the coded speech waveform by determining the stochastic component of the stochastic code vectors in the fixed codebook 16. Index the address for the set. The fixed codebook gain parameter 26 indicates how strong the energy from the coded speech is. The fixed codebook gain parameter 26 includes an output vector from the fixed codebook 16 and a multiplier (mixer) 2.
Multiplied at 7 to provide gain to the fixed codebook data. The long-term prediction block 18, well known in the art, receives a delay parameter 28, which is used to determine the pitch period of the speech waveform. The long-term prediction block 18 also receives a gain parameter 30 to determine the amount of pitch, as is well known in the art.

The short-term prediction LPC synthesis filter 20 utilizes the short-term parameters obtained from the linear prediction coding information to determine the spectral envelope of the coded speech, as is known in the art. predictor). The linear predictive coding filter coefficients are stored in the LDQ LSP reconstruction codebooks 22a to 22a.
Based on the normalized delta quantized data at 22c, it is transformed from the bound ordered set of line spectrum pair parameters and represented by this set. However, it is understood that the disclosed invention is also applicable to non-CELP decoders and other suitable encoding and / or decoding devices.

The normalized delta quantized data in the segmented LSP reconstruction codebooks 22 a to 22 c is used to determine (reconstruct) the absolute value of the line spectrum pair parameter value required for the short-term prediction LPC synthesis filter 20. Used.

[0021] With reference to FIGS. 3 and 4, proper normalization delta quantized data, △ omega _l (see FIG. 5) entry, as shown in block 50, determines a valid LSP parameters each entry Are determined for the split vector LSP reconstruction codebooks 22a-22c so that they can be used to This is performed by the computer based on the training speech frames and other suitable information to be decoded. To avoid the effect of pruning, the delta-encoded LSP is normalized to a range of the constrained space, such as the effective dynamic range. That is, the codebook entry represents the ratio of the distance between the lower bound and the upper bound.
The lower bound can be set to a previously obtained quantized value of a smaller LSP from the same speech frame, or 0.0, the lowest possible value. Similarly, the upper bound can be set to a previously acquired quantized value of a larger LSP from the same speech frame, or 0.5, the maximum possible. This can be expressed as:

[0022]

(Equation 2)

Here, ω _i is the ith non-quantized LSP, ω _H is the upper limit, and ω _L is the lower limit.

At block 52, the normalized delta quantized data is stored in the split vector codebook 22 a-22 c of the decoder 10. This can be done at the time of decoder manufacture, on site request, or at any suitable time. Decoder 10 receives speech parameters 14 for the spectral information associated with the original speech, as shown in block 54. LSP Reconstructor 34
Is based on the NDQ index parameter 32 based on the split vector LS
Extract code vectors from the P reconstruction codebooks 22a to 22c. Code vectors from the split vector LSP reconstruction codebooks 22a-22c include a representation of the estimated LSP parameters. This is indicated by block 56. LS
P reconstructor 34 reconstructs the LSP parameters based on the normalized delta quantized data from codebooks 22a-22c, as shown in block 58.

Utilizing the above relationship, the LSP (ω _i hat) is reconstructed from the codebook value (△ ω _i hat) and the previously obtained lower and upper bounds by:

[0026]

[Equation 3]

The reconstruction of the LSP for each frame is such that the desired lower and upper limits are
Should be performed in the order in which they are available. An implicit point in this method is the fact that at least the first LSP to be reconstructed must be assigned fixed lower and upper limits independent of the frame LSP data.
For example, these fixed values for the first LSP to be reconstructed can be set to a minimum possible value of 0.0 and a maximum possible value of 0.5, which sets the absolute LSP value to the corresponding normalized delta This corresponds to setting to the LSP value. After reconstruction, LSP
The parameters are used by LSP / LPC converter 36 to determine linear predictive coding coefficients, as is well known in the art. Next, the short-term prediction LPC synthesis filter 20 evaluates the next speech parameter, as shown in block 60.

FIG. 5 illustrates one method for generating normalized delta quantized data for split vector codebooks 22 a-22 c. For a parsed frame number by using the distribution of the LSP value omega _i, the user uses the computer, can generate all possible values of △ omega _i for training set. For example, audio to be expected from the encoders, using a model of a frame of music or other data, can be generated distribution range for each △ omega _i. This is indicated by block 70. Next, the computer evaluates the values to obtain an equal error probability distribution for each quantization period, as shown in block 72. Next, the computer takes a distribution average for each quantization level, and sets this average as (△ ω _i hat), thereby split vector codebooks 22a to 22a-2.
2c to index the valid LSP parameters of each split vector codebook, as shown in block 74.

It has been found that the preferred system and method should include utilizing both absolute and normalized delta quantized data. Although the above normalization ensures that all code vectors are valid for any target vector, in another embodiment, the absolute LSP value is used to provide further improved spectral quantization. , The split vector NDQ LSP reconstruction codebook 22
a to 22c.

Referring to FIG. 6, hybrid reconstructed codebooks 80 a to 80 c are split vector codebooks. Codebooks 80a and 80c store at least two split vector segments (groups) that contain absolute split vector code source data, such as absolute LSP values. Codebook 80b stores at least one segment of reconstructed code data representing a normalized delta quantization of a set of order parameters, such as LSP parameters. Reconstructed code data representing the normalized delta quantization of the set of order parameters is optimized based on the set of training data. Hybrid reconstructed codebooks 80a-80c can provide advantages over known lattice codebook structures and the structure shown in FIG. 2 by employing at least two absolute vectors.

[0031] The LSP reconstructor 82 may generate an order parameter based on the absolute split vector code source data and at least two sets of reconstructed code data representing a normalized delta quantization of the set of order parameters. Reconstruct the set of LSP
The reconstructor utilizes at least two segments of the absolute split vector code source data and at least one segment of the normalized delta quantized data to normalize for use by the LSP / LPC converter. Convert the quantized data to absolute quantized data. As described above, the reconstructed code data representing the normalized delta quantization of the set of ordinal parameters is the distance between the two absolute quantized line spectrum pair parameter values or other previously obtained upper and lower limits. Contains the split vector codebook value quantized to represent a percentage.

As an example, the codebook 80a can contain a codebook entry for quantizing one segment or group containing LSP1-3, the codebook 80c can contain LSP7-10, 80b is LSP
4 to 6 can be accommodated. Therefore, codebooks 80a and 80c are absolute LSPs.
Contains the value, while codebook 80b contains the normalized delta quantized data. Furthermore, when calculating the normalized delta quantized data used in deriving the codebook values for codebook 80b, the lower limit omega _L is set equal to the quantization value of LSP3, upper omega _H is the LSP7 Set to the quantization value. From FIG. 1, it can be seen that the distribution of LSP3 and LSP7 has little overlap. Therefore, the loss due to pruning the codebook 80c is small. Use the normalized delta quantization value for quantizing codebook 80c, determined by setting the lower limit to the LSP3 quantization value and the upper limit to the maximum possible value of 0.5. Thus, it is possible to substantially eliminate this pruning loss, but the loss due to warping in this case is greater than the gain due to blocking pruning, so that codebook 80c has an absolute LSP Contains the value.

It can be seen from FIG. 1 that for codebook 80b, there is considerable overlap in the distribution of LSP6 and LSP7 and the distribution of LSP3 and LSP4. Thus, when absolute LSP values are used for codebook 80, there is considerable loss due to pruning. Utilizing the normalized delta quantized data in codebook 80b provides a coding gain due to blocking pruning loss that is greater than the loss due to warping.

In another example, 14 LPCs per frame are available to represent coarse spectral information. The set of 14 LPCs has 14 LPCs in the range of 0.0 to 0.5.
It is converted to an ordered set of SPs. Instead of three codebooks, a four-segment split vector quantizer with four codebooks is used. The first codebook contains entries representing the absolute values of LSP1 to LSP1. The second codebook contains entries representing the absolute values of LSPs 11 to 14. The third codebook contains entries representing normalized delta LSP values derived from LSPs 7-10. When calculating these normalized delta quantization values, the lower limit is LSP
3 is set equal to the quantization value, and the upper limit is set equal to the quantization value of LSP11. The fourth codebook contains entries representing normalized delta LSP values derived from LSPs 4-6. When calculating these normalized delta LSP values, the lower limit is set equal to the quantized value of LSP3, and the upper limit is set to the quantized value of LSP7.

It should be understood that implementation of other variations and modifications of the invention in various aspects will be apparent to those skilled in the art, and that the invention is not limited to the particular embodiments described. For example, the present systems and methods can be applied to scalars, vectors, matrices or other suitable quantizers having various dimensions and segmentations. In addition, there are other similar normalization methods that have the effect of blocking codebook pruning and can therefore be used to achieve a similar improvement in the performance of split vector quantization of ordered sets. For example, the normalized LSP can be calculated according to the following equation:

[0036]

(Equation 4)

[0037] Here, a _{ω L <ω i <ω H} , an _{ω L ≦ ω r ≦ ω H} , also △ omega _i is the normalized LS P values, omega _i is the absolute LSP value, ω _H and ω _L are the upper and lower limits obtained before satisfying the above constraint, ω _r is the previously obtained reference value satisfying the above constraint, and the sign (sign) is The same polarity is set a priori. In this case, the absolute quantized LSP value is obtained from the quantized value of △ ω _i (or (△ ω _i hat)) according to the following equation.

[0038]

(Equation 5)

It is also understood that the combination of absolute and normalized delta quantization values will vary depending on the statistics of the data being quantized. Accordingly, the present invention is intended to cover all modifications, variations, or equivalents, within the spirit and scope of the basic principles disclosed and claimed herein.

[Brief description of the drawings]

FIG. 1 is a graph showing the line spectrum pair parameter distribution of a number of analyzed speech frames.

FIG. 2 is a block diagram schematically illustrating a speech decoder incorporating one embodiment of the present invention in a wireless communication system.

FIG. 3 is a diagram schematically illustrating an example of a codebook structure including a normalized quantization according to an embodiment of the present invention.

FIG. 4 is a flow diagram that schematically illustrates a method for performing split vector quantization to determine spectral parameters, according to one embodiment of the present invention.

FIG. 5 illustrates a split vector line spectrum according to one embodiment of the present invention.
FIG. 7 is a flowchart schematically illustrating a method of determining normalized delta quantized data for a pair codebook.

FIG. 6 is a block diagram schematically illustrating a speech decoder incorporating another embodiment of the present invention in a wireless communication system, wherein a codebook memory stores absolute LSP values and normalized delta quantized data. Accommodates both combinations.

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Mark A. Jashiuk 6221, North Melvina Avenue, Chicago, Illinois, USA (72) Inventor Aaron M. Smith, Jackson Lane 64, Streamwood, Illinois, USA F term (reference) 5D045 CA03 CC07 5J064 BA12 BA13 BB03 BC02 BC14 BC17 BD02

Claims (10)

[Claims] 1. A method for performing split vector quantization used in examining a set of order parameters for encoding, comprising: a reconstruction representing normalized delta quantized data for examining a set of order parameters. Storing a split vector reconstructed code source containing code data, wherein each entry of the reconstructed code data corresponds to valid data for examining the set of order parameters, and Facilitating parameter reconstruction; and the received spectrum based on reconstruction code data representing a normalized delta quantization of the set of ordinal parameters from the split vector reconstruction code source. Reconstructing the parameters. 2. The split vector reconstructed code source is a split vector codebook, and the set of order parameters is obtained from the normalized delta quantized data. The described method. 3. The reconstructed code data representing a normalized delta quantization of the set of order parameters, wherein the reconstructed code data is quantized to represent a percentage distance between a lower limit and an upper limit. 3. The method of claim 2, including a value. 4. The step of reconstructing received parameters comprises dynamically reconstructing line spectrum pair (LSP) parameters based on the reconstructed code data representing a normalized delta quantization of the set of order parameters. The method of claim 1, wherein the received parameters are reconstructed using linear predictive coding coefficients based on dynamically generated LSP parameters. 5. Pointing each vector entry of the split vector reconstructed code source to facilitate selection of a vector entry from a different split vector in the split vector reconstructed code source. The method of claim 1, further comprising: 6. The method of claim 1, further comprising the step of operatively receiving encoded spectral parameters from a speech encoder. 7. The method of claim 1, wherein the set of order parameters comprises line spectrum pair values. 8. The method of claim 1, further comprising storing at least two groups of absolute split vector code source data, wherein reconstructing the received parameters. Is an absolute split
Reconstructing received parameters based on two sets of vector code source data and the reconstructed code data representing a normalized delta quantization of the set of order parameters. The method of claim 1, wherein 9. The reconstructed code data representing a normalized delta quantization of the set of order parameters, wherein the reconstructed code data is quantized to represent a distance percentage between two quantized LSP values and an absolute quantized value. 3. The method of claim 2 including a vector codebook value. 10. An encoding system for performing split vector quantization for use in determining a set of order parameters, the reconstruction system representing normalized delta quantized data for determining the set of order parameters. Means for storing a split vector reconstructed code source containing code data, wherein each entry of the reconstructed code data corresponds to valid data for determining the set of order parameters, and Means for facilitating and storing the reconstructed spectral parameters; and operably coupled to the storing means for performing a normalized delta quantization of the set of ordinal parameters from the split vector reconstructed code source. Means for reconstructing the received parameters based on the represented reconstruction code data; A system characterized by being constituted.