AU745837B2

AU745837B2 - Method and apparatus for producing an aproximation to a sequence of vectors

Info

Publication number: AU745837B2
Application number: AU71458/00A
Authority: AU
Inventors: Christian Herbert Ritz; Burnett Ian Shaw; Mark Mclean Thomson
Original assignee: University of Wollongong; Motorola Australia Pty Ltd
Current assignee: University of Wollongong; Motorola Solutions Australia Pty Ltd
Priority date: 1999-11-12
Filing date: 2000-11-06
Publication date: 2002-04-11
Anticipated expiration: 2020-11-06
Also published as: AU7145800A

Description

P/00/011 Regulation 3.2

AUSTRALIA

Patents Act 1990 *~c

S

ORIGINAL

COMPLETE SPECIFICATION STANDARD PATENT Invention Title: "METHOD AND APPARATUS FOR PRODUCING AN APPROXIMATION TO A SEQUENCE OF VECTORS" The following statement is a full description of this invention, including the best method of performing it known to me/us: I 1 CR1058AC METHOD AND APPARATUS FOR PRODUCING AN APPROXIMATION TO A SEQUENCE OF VECTORS FIELD OF THE INVENTION The invention relates to a method and apparatus for producing an approximation to a sequence of vectors, especially, though not exclusively, so 10 as to improve the performance of a class of methods used for compressing digitised speech for storage or for transmission over digital communication ooooo channels.

BACKGROUND OF THE INVENTION Most speech compression systems involve analysing multiple segments 00*00: o of a speech waveform, in order to extract from each segment a set of o:oso: parameters. Each set of parameters can be regarded as a vector, and the analysis process as a whole produces a sequence of vectors. These vectors are 20 referred to as true vectors.

For the purposes of data compression, it is desirable to construct a second sequence of vectors which approximates the sequence of true vectors, and which can be represented by a smaller number of parameters than those contained in the sequence of true parameters. This smaller set of parameters is referred to as representation parameters. A sequence of approximating vectors can be represented by a set of representation parameters if the representation parameters can be used as inputs to some known process that can generate the approximating vectors.

BRIEF DESCRIPTION OFTHE INVENTION It is an object of the present invention to provide a method and apparatus for producing an approximation to a sequence of vectors.

Accordingly, in one aspect, the invention provides a method of producing an approximation to a sequence of vectors, the method comprising the steps of receiving a sequence of input vectors, wherein each vector comprises a plurality of parameters; generating from the sequence of input vectors a plurality of sequences of true split vectors; determining for each sequence of true split vectors a set of true target vectors and associated true event functions and; generating for each true target vector an approximate target oooo o 15 vector, wherein each approximate target vector has the same number of elements as the associated true target vector; generating for each true event function value an approximate event function value; generating for each sequence of true split vectors a sequence of 20 approximate split vectors, wherein an approximate split vector is obtained by summing approximate split vector components; and generating a sequence of approximate input vectors wherein an approximate input vector comprises elements from a plurality of corresponding approximate split vectors.

Preferably, each true split vector may comprise a subset of the parameters contained in a corresponding input vector.

Suitably, a true event function may comprise distinct values associated with each true split vector.

Preferably, one or more true event functions and true target vectors may be determined such that a corresponding sequence of model split vectors approximate the sequence of true split vectors.

Suitably, a model split vector can be obtained by summing model split vector components.

Preferably, a model split vector component may be a product of a true target vector and a true event function value associated with the true target vector and with the model split vector.

Suitably, an approximate split vector component may be a product of a quantised target vector and a quantised event function value associated with the quantised target vector and with the approximate split vector.

**According to a second aspect, the invention provides an apparatus for producing an approximation to a sequence of vectors, the apparatus comprising: an input terminal for receiving a sequence of input vectors, wherein each input vector comprises a plurality of parameters; a splitter for generating from the sequence of input vectors a 15 plurality of sequences of true split vectors; an analyser for determining for each sequence of true split vectors a set of true target vectors and associated true event functions; a first codec for generating for each true target vector an approximate target vector, wherein each approximate target vector has the same number of elements as the associated true target vector; a second codec for generating for each true event function value an approximate event function value; a synthesiser for generating for each sequence of true split vectors a sequence of approximate split vectors; and a combiner for generating a sequence of output vectors wherein an output vector comprises elements from a plurality of corresponding approximate split vectors.

Suitably, an approximate split vector may be obtained by summing approximate split vector components.

Preferably, an approximate split vector component may be a product of quantised target vector and a quantised event function value associated with the quantised target vector and with the approximate split vector ooo.

BRIEF DESCRIPTION OF THE DRAWINGS An embodiment of the invention will now be more fully described, by way of example, with reference to drawings, of which: FIG. 1 shows a block diagram of an apparatus used to produce an approximation to a sequence of vectors; and FIG. 2 shows a flowchart of a method for producing an approximation to a sequence of vectors utilising the apparatus of FIG. 1.

DETAILED DESCRIPTION OF THE DRAWING Thus, the overall scheme of the invention according to one embodiment is shown in FIG. 1.

As shown in FIG. 1, multidimensional input vectors are provided from an input terminal 100 to a splitter 101. The input terminal 100 receives a sequence of multidimensional input vectors and the splitter 101 produces a plurality of sequences of true split vectors from the sequence of multidimensional input vectors. In a preferred embodiment, the elements of each true split vector are the same as a subset of the elements of a corresponding input vector. Moreover each element of a particular input vector corresponds to only one element of a corresponding true split vector. This procedure of receiving a sequence of input vectors where each of the vectors comprises a plurality of parameters is shown at steps 201 and 202 in FIG. 2.

The sequences of true split vectors are provided from the splitter 101 to an analyser 102. The analyser 102 calculates true target vectors and true event functions for each sequence of true split vectors. This procedure of generating sequences of true split vectors from the sequence of input vectors is shown at step 203 in FIG. 2. Each of the true split vectors comprises a subset of the parameters contained in the corresponding input vector, The true target vectors and the true event functions are determined in such a way that a sequence of model vectors computed using the following expression approximate the sequence of true split vectors as closely as possible according to some criterion: Yji ajkiXjk k where Yj,i is a model vector corresponding to the ith true split vector of the.jth sequence of true split vectors, Xjk is the kth true target vector associated with the.jth sequence of true split vectors, and aj,k, i is the ith event function value associated with the .jth sequence of true split vectors and the kth true target vector.

Various methods of determining target vectors and event functions for a sequence of vectors are known. One example is the method described in "Efficient coding of LPC parameters by Temporal Decomposition", by B. S.

Atal, in Proceedings of the International Conference on Acoustics, Speech and Signal Processing" (Boston, 1983), pp. 81-84.

The true target vectors are fed into a first codec 103, which produces approximate target vectors. Preferably, the first codec 103 comprises a quantiser, which converts true target vectors to binary codewords, a codebook which converts binary codewords to approximate target vectors, and a communication system which conveys binary codewords from the quantiser to the codebook. This procedure is shown at step 204 in FIG. 2.The true event functions are fed into a second codec 104, which produces approximate event functions. Preferably, the second codec 104 comprises a quantiser, which converts the true event function values to binary codewords, a codebook which maps the binary codewords to approximate event function values, and a communication system which conveys the binary codewords from the quantiser to the codebook. This procedure of determining for each sequence of true split vectors a set of true target vectors and associated true event functions is shown at step 204 in FIG. 2. In this regard, a true event function comprises distinct values associated with each true split vector and true event functions and true target vectors are determined such that a corresponding sequence of model split vectors approximate the sequence of true split vectors. A model split vector is obtained by summing model split vector components, and a model split vector component is a product of a true target vector and a true event function value associated with the true target vector and with the model split vector; Methods of converting vectors or numerical values to codewords and codewords to approximate vectors or numerical values, respectively, and methods for conveying codewords using a communication system are well 20 known, as described, for example in, Introduction to Communication Systems, by F.G. Stremler (Addison-Wesley, 1990).

The approximate target vectors together with the approximate event functions are fed into a synthesiser 105. The synthesiser 105 calculates a sequence of approximate split vectors for each set of approximate target vectors and associated approximate event functions. An approximate split vector, Yj,i corresponding to the ith true split vector of the jth sequence of true split vectors is computed as Yj,i jk,jk k where jj,k is the kth approximate target vector associated with the ith sequence of approximate split vectors, and a.,k,i is the ith event function value associated with the jth sequence of true split vectors and the kth approximate target vector. This procedure of generating for each true target vector an approximate target vector and generating for each true event function value an approximate event function value is shown at step 205 in FIG. 2. In this regard, each approximate target vector has the same number of elements as the associated true target vector; The sequences of approximate split vectors are fed into a combiner 106, which produces a sequence of multidimensional output vectors. In the preferred 1 embodiment, each element of an output vector is the same as a corresponding 10 element of a corresponding approximate split vector. This procedure of ot.o generating for each sequence of true split vectors a sequence of approximate i' splitvectors and then generating a sequence of approximate input vectors is shown at step 206 in FIG. 2. An approximate split vector is obtained by *go.

summing approximate split vector components, and an approximate split vector component is a product of a quantised target vector and a quantised event function value associated with the quantised target vector and with the approximate split vector. Furthermore, an approximate input vector comprises elements from a plurality of corresponding approximate split vectors.

0:.:00 It will be appreciated that although only one particular embodiment has been described here in detail, various modifications and improvements can be made by a person skilled in the art without departing from the scope of the present invention.

Claims

1. A method of producing an approximation to a sequence of vectors, the method comprising the steps of: receiving a sequence of input vectors, wherein each vector comprises a plurality of parameters; generating from the sequence of input vectors a plurality of sequences of true split vectors; determining for each sequence of true split vectors a set of true target vectors and associated true event functions; generating for each true target vector an approximate target •vector, wherein each approximate target vector has the same number of elements as the associated true target vector; generating for each true event function value an approximate 15 event function value; generating for each sequence of true split vectors a sequence of approximate split vectors, wherein an approximate split vector is obtained by summing approximate split vector components; and generating a sequence of approximate input vectors wherein an approximate input vector comprises elements from a plurality of corresponding approximate split vectors.

2. A method as claimed in claim 1, wherein each true split vector comprises a subset of the parameters contained in a corresponding input vector.

3. A method as claimed in claim 1, wherein a true event function comprises distinct values associated with each true split vector.

4. A method as claimed in claim 1, wherein one or more true event functions and true target vectors is determined such that a corresponding sequence of model split vectors approximate the sequence of true split vectors.

A method as claimed in claim 4, wherein a model split vector is obtained by summing model split vector components.

6. A method as claimed in claim 5, wherein a model split vector component is a product of a true target vector and a true event function value associated with the true target vector and with the model split vector.

7. A method as claimed in claim 1, wherein an approximate split C ~vector component is a product of a quantised target vector and a quantised %o 10 event function value associated with the quantised target vector and with the 0:o0 approximate split vector. 0S

8. An apparatus for producing an approximation to a sequence of vectors, the apparatus comprising: an input terminal for receiving a sequence of input vectors, soots: wherein each input vector comprises a plurality of parameters; a 4 a splitter for generating from the sequence of input vectors a plurality of sequences of true split vectors; an analyser for determining for each sequence of true split se ease 20 vectors a set of true target vectors and associated true event functions; a first codec for generating for each true target vector an approximate target vector, wherein each approximate target vector has the same number of elements as the associated true target vector; a second codec for generating for each true event function value an approximate event function value; a synthesiser for generating for each sequence of true split vectors a sequence of approximate split vectors; and a combiner for generating a sequence of output vectors wherein an output vector comprises elements from a plurality of corresponding approximate split vectors.

9. An apparatus as claimed in claim 8, wherein each true split vector comprises a subset of the parameters contained in a corresponding input vector.

10. An apparatus as claimed in claim 8, wherein a true event function comprises distinct values associated with each true split vector.

11. An apparatus as claimed in claim 10, wherein one or more true event functions and true target vectors are determined such that a 10 corresponding sequence of model split vectors approximate the sequence of true split vectors.

12. An apparatus as claimed in claim 11, wherein a model split Svector is obtained by summing model split vector components.

13. An apparatus as claimed in claim 12, wherein a model split .o,.oi vector component is a product of a true target vector and a true event function 0*0*0* value associated with the true target vector and with the model split vector. 20

14. An apparatus as claimed in claim 8, wherein an approximate split vector is obtained by summing approximate split vector components.

An apparatus as claimed in claim 14, wherein an approximate split vector component is a product of a quantised target vector and a quantised event function value associated with the quantised target vector and with the approximate split vector.

16. A method of producing an approximation to a sequence of vectors substantially as hereinbefore described with reference to the accompanying drawings.

17. An apparatus for producing an approximation to a sequence of vectors substantially described with reference to the accompanying drawings.