Speech coding system having codebook storing differential vectors between each two adjoining code vectors

• US 5,323,486 A
• Filed: 05/14/1992
• Issued: 06/21/1994
• Est. Priority Date: 09/14/1990
• Status: Expired due to Term

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Tanaka, Yoshinori, Taniguchi, Tomohiko, Johnson, Mark, Kurihara, Hideaki, Sakai, Yoshihiro, Ohta, Yasuji
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Doerrler, Michelle
• Application Number:
  US07/856,221
• Time in Patent Office: 768 Days
• Field of Search: 381/29-41, 395/2.31, 395/2.29
• US Class Current: 704/222
• CPC Class Codes: G10L 19/12 the excitation function being a code excitation, e.g. in code excited linear prediction [CELP] vocoders
• CPC Class Codes: G10L 2019/0002 Codebook adaptations
• CPC Class Codes: G10L 2019/0007 Codebook element generation
• CPC Class Codes: G10L 2019/0013 Codebook search algorithms
• CPC Class Codes: G10L 2019/0014 Selection criteria for distances
• CPC Class Codes: G10L 25/06 the extracted parameters being correlation coefficients

Abstract

A speech coding system is provided where input speech is coded by finding via an evaluation computation a code vector giving a minimum error between reproduced signals obtained by linear prediction analysis filter processing, simulating speech path characteristics, on code vectors successively read out from a noise codebook storing a plurality of noise trains as code vectors and an input speech signal and by using a code specifying the code vector. In the speech coding system, the noise codebook includes a delta vector codebook which stores an initial vector and a plurality of delta vectors having difference vectors between adjoining code vectors. In addition, provision is made in the computing unit for the evaluation computation of a cyclic adding unit for cumulatively adding the delta vectors to virtually reproduce the code vectors.

Claims

1. A speech coding system coding input speech by evaluation computation producing a single code vector providing a minimum error between an input speech signal and reproduced signals generated by a linear prediction analysis filter, the linear prediction analysis filter using code vectors successively read from a noise codebook storing a plurality of noise trains as the code vectors and a code specifying the single code vector, said speech coding system comprising:

said noise codebook, connected to the linear prediction analysis filter and including a delta vector codebook storing an initial vector and a plurality of delta vectors produced using differential vectors determined between adjoining code vectors for all of the code vectors, and said plurality of delta vectors being cyclically added to reproduce the code vectors.

2. A speech coding system as set forth in claim 1, wherein said plurality of delta vectors comprise N dimensional vectors each comprised of N number (N being a natural number of at least 2) of time-series sample data, and several of the N number of time-series sample data are significant data, and others of the N number of time-series sample data are sparsed vectors comprised of data 0.

3. A speech coding system coding input speech by evaluation computation producing a single code vector providing a minimum error between an input speech signal and reproduced signals generated by a linear prediction analysis filter, the linear prediction analysis filter using code vectors successively read from a noise codebook storing a plurality of noise trains as the code vectors and a code specifying the single code vector, said speech coding system comprising:

said noise codebook, connected to the linear prediction analysis filter and including a delta vector codebook storing an initial vector and a plurality of delta vectors produced using differential vectors determined between adjoining code vectors for all of the code vectors, and said plurality of delta vectors being cyclically added to reproduce the code vectors,

wherein said plurality of delta vectors comprise N dimensional vectors each comprised of N number (N being a natural number of at least

     2) of time-series sample data, and several of the N number of time-series sample data are significant data, and others of the N number of time-series sample data are sparsed vectors comprised of data 0, and

wherein the code vectors in the noise codebook are rearranged as rearranged code vectors so that the differential vectors determined between the adjoining code vectors become smaller, and wherein the differential vectors between the adjoining code vectors are determined for the rearranged code vectors, and the sparsed vectors are obtained using the differential vectors.

4. A speech coding system coding input speech by evaluation computation producing a single code vector providing a minimum error between an input speech signal and reproduced signals generated by a linear prediction analysis filter, the linear prediction analysis filter using code vectors successively read from a noise codebook storing a plurality of noise trains as the code vectors and a code specifying the single code vector, said speech coding system comprising:

said noise codebook, connected to the linear prediction analysis filter and including a delta vector codebook storing an initial vector and a plurality of delta vectors produced using differential vectors determined between adjoining code vectors for all of the code vectors, and said plurality of delta vectors being cyclically added to reproduce the code vectors; and

computing means for performing the evaluation computation, and said computing means including cyclic adding means for performing cyclic addition on said plurality of delta vectors.

5. A speech coding system as set forth in claim 4, wherein said cyclic adding means comprises:

adding unit means having inputs for adding the plurality of delta vectors and outputting an add signal; and

delay unit means for delaying the add signal output from the adding unit means and outputting a delayed signal being input to one of the inputs of the adding unit means, and

wherein previous computation results are held in said delay unit means and a next delta vector is used as the input to said adding unit means, and the evaluation computation is cumulatively updated.

6. A speech coding system coding input speech by evaluation computation producing a single code vector providing a minimum error between an input speech signal and reproduced signals generated by a linear prediction analysis filter, the linear prediction analysis filter using code vectors successively read from a noise codebook storing a plurality of noise trains as the code vectors and a code specifying the single code vector, said speech coding system comprising:

said noise codebook, connected to the linear prediction analysis filter and including a delta vector codebook storing an initial vector and a plurality of delta vectors produced using differential vectors determined between adjoining code vectors for all of the code vectors, and said plurality of delta vectors being cyclically added to reproduce the code vectors,

wherein the plurality of delta vectors include (L-1) types of delta vectors arranged in a tree-structure having a peak, where L is a total number of layers comprising the tree-structure with the initial vector located at the peak.

7. A speech coding system as set forth in claim 6, wherein the (L-1) types of delta vectors are one of successively added to and successively subtracted from the initial vector for each of the layers to virtually reproduce (2^L -1) types of code vectors.

8. A speech coding system as set forth in claim 7,

wherein the code vectors include 2^L types of code vectors, and

wherein zero vectors are added to the (2^L -1) types of code vectors to reproduce 2^L types of reproduced code vectors of the same number as the 2^L types of code vectors stored in said noise codebook.

9. A speech coding system as set forth in claim 7,

wherein the code vectors include 2^L types of code vectors, and

wherein one of the code vectors generated by multiplying the initial vector by -1 is added to the (2^L -1) types of code vectors to reproduce the 2^L types of reproduced code vectors of the same number as the 2^L types of code vectors stored in said noise codebook.

10. A speech coding system as set forth in claim 6, further comprising computing means for performing the evaluation computation, and said computing means including cyclic adding means for performing cyclic addition on said plurality of delta vectors.

11. A speech coding system as set forth in claim 10, wherein said evaluation computation performed by said computing means includes a cross correlation computation of a cross correlation and a linear prediction analysis filter computation of an analysis filter computation output comprised of a first recurrence equation using a previous analysis filter computation output from a previous layer and one of the plurality of delta vectors, whereby the cross correlation computation is performed using a second recurrence equation.

12. A speech coding system as set forth in claim 11,

wherein said evaluation computation performed by said computing means includes an auto correlation computation of an auto correlation, and

wherein the analysis filter computation output is comprised of the first recurrence equation using the previous analysis filter computation output from the previous layer and the one of the plurality of delta vectors, whereby the auto correlation computation is performed using an L number of auto correlations of the analysis filter computation output computed from the initial vector, a filter computation output of the (L-1) types of delta vectors and (L² -1)/2 types of cross correlations using the analysis filter computation output.

13. A speech coding system as set forth in claim 6, wherein an order of the initial vector and said (L-1) types of delta vectors in the tree-structured is rearranged responsive to properties of the input speech.

14. A speech coding system as set forth in claim 13, wherein the initial vector and the (L-1) types of delta vectors are stored and rearranged in frames responsive to filter properties of the linear prediction analysis filter performing the linear prediction analysis filter computation, and one of the evaluation computations.

15. A speech coding system as set forth in claim 14, wherein a first power of each of said reproduced signals generated by the linear prediction analysis filter is evaluated by said evaluation computation and the code vectors are rearranged in a new order successively from one of the code vectors corresponding to one of the reproduced signals with the first power most increased compared with a second power of the one of the code vectors determined before the reproduced signals are generated.

16. A speech coding system as set forth in claim 15, wherein said initial vector and the (L-1) delta vectors are transformed in advance to be mutually orthogonal with each other after the filter processing, and the initial vector and the plurality of delta vectors in the delta vector codebook are uniformly distributed on a hyper plane.

17. A speech coding system as set forth in claim 15, wherein a magnitude of the first power is compared with a normalized power obtained by normalization of each first power.

18. A speech coding system as set forth in claim 13, wherein said code specifying the single code vector is specified so that a first intercode distance belonging to higher layers in the tree-structure becomes greater than a second intercode distance belonging to lower layers.

19. A noise codebook storing noise trains as code vectors in a speech coding system, comprising:

a delta vector codebook storing an initial vector and delta vectors produced from differences determined between the code vectors, and said initial and delta vectors being used to reproduce the code vectors.

20. A noise codebook storing noise trains as code vectors in a speech coding system, comprising:

a delta vector codebook storing an initial vector and delta vectors produced from differences determined between the code vectors, and said initial and delta vectors being used to reproduce the code vectors,

wherein the code vectors C_i, i being a first integer between 0 and (m-1), and m being a second integer representing a number of the noise trains stored in the noise codebook, are generated using said delta vectors Δ

C_i according to;

##EQU9##

21. A noise codebook storing noise trains as code vectors in a speech coding system, comprising:

a delta vector codebook storing an initial vector and delta vectors produced from differences determined between the code vectors, and said initial and delta vectors being used to reproduce the code vectors,

wherein the code vectors are generated by computing and cyclically adding the delta vectors.

22. A noise codebook storing noise trains as code vectors in a speech coding system, comprising:

a delta vector codebook storing an initial vector and delta vectors produced from differences determined between the code vectors, and said initial and delta vectors being used to reproduce the code vectors,

wherein a linear production analysis filter is used to compute powers of said initial and delta vectors, and

wherein said initial and delta vectors are stored in an order in said delta vector codebook based on said powers.

23. A noise codebook storing noise trains as code vectors in a speech coding system, comprising:

a delta vector codebook storing an initial vector and delta vectors produced from differences determined between the code vectors, and said initial and delta vectors being used to reproduce the code vectors,

wherein said delta vector codebook stores said initial vector and (L-1) types of said delta vectors based on a tree-structure having stages, L being a first number of said stages in said tree-structure.

24. A noise codebook as set forth in claim 23, wherein the code vectors C_i, i being a first integer between 0 and (m-1), and m being a second integer representing a second number of the noise trains stored in the noise codebook, are generated using said delta vectors Δ

C_i according to;

##EQU10##

25. A noise codebook as set forth in claim 23,

wherein said stages include high and low stages, and

wherein a first group of said initial and delta vectors having first intercode distances are stored in said high stages, and a second group of said initial and delta vectors having second intercode distances are stored in said low stages, and said first intercode distances being greater than said second intercode distances.

26. A method of storing noise trains as code vectors in a noise codebook included in a speech coding system, comprising the steps of:

(a) storing an initial vector in a delta vector codebook included in the noise codebook; and

(b) storing delta vectors determined from differences between the code vectors in the delta vector codebook, where the initial and delta vectors are used to reproduce the code vectors.

27. A method of storing noise trains as code vectors in a noise codebook included in a speech coding system, comprising the steps of:

(a) storing an initial vector in a delta vector codebook included in the noise codebook;

(b) storing delta vectors determined from differences between the code vectors in the delta vector codebook, where the initial and delta vectors are used to reproduce the code vectors; and

(c) generating the code vectors C_i, i being a first integer between 0 and (m-1), and m being a second integer representing a number of the noise trains stored in the noise codebook, according to;

##EQU11##

28. A method of storing noise trains as code vectors in a noise codebook included in a speech coding system, comprising the steps of:

(a) storing an initial vector in a delta vector codebook included in the noise codebook;

(b) storing delta vectors determined from differences between the code vectors in the delta vector codebook, where the initial and delta vectors are used to reproduce the code vectors; and

(c) generating the code vectors by computing and cyclically adding the delta vectors.

29. A method of storing noise trains as code vectors in a noise codebook included in a speech coding system, comprising the steps of:

(a) storing an initial vector in a delta vector codebook included in the noise codebook;

(b) storing delta vectors determined from differences between the code vectors in the delta vector codebook, where the initial and delta vectors are used to reproduce the code vectors;

(c) computing powers of the initial and delta vectors; and

(d) re-storing the initial and delta vectors in the delta vector codebook based on the powers computed in said computing step (c).

30. A method of storing noise trains as code vectors in a noise codebook included in a speech coding system, comprising the steps of:

(a) storing an initial vector in a delta vector codebook included in the noise codebook; and

(b) storing delta vectors determined from differences between the code vectors in the delta vector codebook, where the initial and delta vectors are used to reproduce the code vectors,

wherein said storing step (a) and said storing step (b) store the initial vector and (L-1) types of the delta vectors based on a tree-structure having stages, where L is a first number of the stages in the tree-structure.

31. A method as set forth in claim 30, further comprising, after said storing step (b), the step of generating the code vectors C_i, i being a first integer between 0 and (m-1), and m being a second integer representing a second number of the noise trains stored in the noise codebook, according to:

##EQU12##

32. A method as set forth in claim 30,

wherein said stages include high and low stages, and

wherein said method further comprises, after said storing step (b), the step of re-storing in the delta vector codebook a first group of the initial and delta vectors having first intercode distances in the high stages, and a second group of the initial and delta vectors having second intercode distances in the low stages, and the first intercode distances being greater than the second intercode distances.