Vector quantizing for lpc parameters

• US 20040015346A1
• Filed: 05/27/2003
• Published: 01/22/2004
• Est. Priority Date: 11/30/2000
• Status: Active Grant

• Current Assignee: Nippon Telegraph and Telephone Corporation, Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Original Assignee: Nippon Telegraph and Telephone Corporation, Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Morii, Toshiyuki, Hiwasaki, Yusuke, Ehara, Hiroyuki, Yasunaga, Kazutoshi, Mano, Kazunori
• Granted Patent: US 7,392,179 B2
• Time in Patent Office: Days
• US Class Current: 704/219
• CPC Class Codes: G10L 19/07 Line spectrum pair [LSP] vocoders
• CPC Class Codes: G10L 2019/0005 Multi-stage vector quantisation
• CPC Class Codes: G10L 2019/0014 Selection criteria for distances
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94

Abstract

The present invention carries out pre-selection on many LPC codevectors stored in an LSF codebook 101 using a weighted Euclidean distortion as a measure and carries out a full-code selection on the LPC codevectors left after the pre-selection using an amount of distortion in a spectral space as a measure. This makes it possible to improve the quantization performance of the LPC parameter vector quantizer and improve the quality of synthesized speech of the speech coder/decoder.

Claims

1. An LPC parameter vector quantization apparatus comprising:

a first calculation section that calculates distortion between a plurality of codevectors read successively from a codebook storing many codevectors and a quantization target using a first distortion evaluation measure;

a pre-selection section that preliminarily selects codevectors that reduce the amount of distortion according to said first distortion evaluation measure from said many codevectors;

a second calculation section that calculates distortion between the preliminarily selected codevectors and said quantization target using a second distortion evaluation measure which is different from said first distortion evaluation measure; and

a final search section that specifies a codevector that minimizes the amount of distortion according to said second distortion evaluation measure from the preliminarily selected codevectors and outputs index information of the specified codevector.

2. The LPC parameter vector quantization apparatus according to claim 1, wherein the quantization target is an LPC parameter that represents short-time spectral envelope information of a speech signal.

3. The LPC parameter vector quantization apparatus according to claim 2, wherein the LPC parameter is an LPC coefficient obtained by carrying out a linear predictive analysis on a speech signal in a processing frame section.

4. The LPC parameter vector quantization apparatus according to claim 2, wherein the LPC parameter is a parameter obtained by converting an LPC coefficient obtained by carrying out a linear predictive analysis on a speech signal of a processing frame section to a parameter in a frequency area.

5. The LPC parameter vector quantization apparatus according to claim 2, wherein the LPC parameter is an LSF parameter obtained by converting an LPC coefficient obtained by carrying out a linear predictive analysis on a speech signal of a processing frame section.

6. The LPC parameter vector quantization apparatus according to claim 2, wherein the LPC parameter is an LSP parameter obtained by converting an LPC coefficient obtained by carrying out a linear predictive analysis on a speech signal of a processing frame section.

7. The LPC parameter vector quantization apparatus according to claim 2, wherein the first distortion evaluation measure used to calculate distortion between the codevectors read successively from the codebook and the quantization target is a Euclidean distortion.

8. The LPC parameter vector quantization apparatus according to claim 2, wherein the first distortion evaluation measure used to calculate distortion between the codevectors read successively from the codebook and the quantization target is a weighted Euclidean distortion.

9. The LPC parameter vector quantization apparatus according to claim 2, wherein the second distortion evaluation measure used to calculate distortion between the preliminarily selected codevectors and the quantization target is a measure to evaluate an amount of distortion in a frequency spectral space.

10. The LPC parameter vector quantization apparatus according to claim 9, wherein the second distortion evaluation measure used to calculate distortion between the preliminarily selected codevectors and the quantization target is cepstrum distortion (CD).

11. The LPC parameter vector quantization apparatus according to claim 9, wherein the second distortion evaluation measure used to calculate distortion between the preliminarily selected codevectors and the quantization target is spectrum distortion (SD).

12. The LPC parameter vector quantization apparatus according to claim 2, wherein the number of codevectors preliminarily selected by the pre-selection section is specified beforehand.

13. The LPC parameter vector quantization apparatus according to claim 2, wherein the pre-selection section presets a threshold as the amount of distortion by the first distortion evaluation measure and preliminarily selects codevectors whereby the amount of distortion by the first evaluation measure is smaller than said set threshold.

14. An LPC parameter decoder comprising:

an input section that inputs index information output by the vector quantization apparatus according to claim 2;

and

a reading section that reads any codevector from the same codebook as the codebook at said vector quantization apparatus and a plurality of codevectors stored in said codebook;

wherein the codevector corresponding to said index information is read from said codebook and the read codevector is output as a decoded LPC parameter.

18. The LPC parameter vector quantization apparatus according to claim 2 that carries out predictive vector quantization.

19. The LPC parameter vector quantization apparatus according to claim 2 that carries out split vector quantization.

20. The LPC parameter vector quantization apparatus according to claim 2 that carries out predictive vector quantization and split vector quantization.

21. The LPC parameter vector quantization apparatus according to claim 2 that carries out third-order MA predictive vector quantization, 2-stage vector quantization and split vector quantization, vector-quantizes an LPC parameter input as a quantization target and outputs an LPC code.

24. An LPC parameter decoder that inputs an LPC code output by the LPC parameter vector quantization apparatus according to claim 21 and generates a decoded LPC parameter based on said input LPC code.

31. A speech coder using the LPC parameter vector quantization apparatus according to claim 2, comprising:

a section that quantizes/codes spectrum characteristics of an input speech signal;

an adaptive excitation codebook that represents a cyclic component of an input speech signal;

a fixed excitation codebook;

a section that synthesizes a synthesized speech signal using excitation vectors generated from said fixed excitation codebook and said adaptive excitation codebook, and said parameters; and

a section that determines the outputs from said fixed codebook and said adaptive codebook so that distortion between an input speech signal and said synthesized speech signal decreases.

32. A speech decoder using the LPC parameter decoder according to claim 14, comprising:

a section that decodes parameters representing spectrum characteristics coded by a speech coder;

a fixed excitation codebook;

an adaptive excitation codebook that represents a cyclic component of a synthesized speech signal; and

a section that decodes excitation vectors determined by said speech coder from said fixed excitation codebook and said adaptive excitation codebook and synthesizes a synthesized speech signal from the decoded excitation vector and said parameter.

33. A speech signal transmission apparatus equipped with a speech coder, said speech coder using the LPC parameter vector quantization apparatus according to claim 2 comprising:

a section that quantizes/codes spectrum characteristics of an input speech signal;

an adaptive excitation codebook that represents a cyclic component of an input speech signal;

a fixed excitation codebook;

a section that synthesizes a synthesized speech signal using excitation vectors generated from said fixed excitation codebook and said adaptive excitation codebook and said parameter; and

a section that determines the outputs from said fixed codebook and said adaptive codebook so that distortion between an input speech signal and said synthesized speech signal decreases.

34. A speech signal reception apparatus equipped with a speech decoder, said speech decoder using the LPC parameter decoder according to claim 14 comprising:

a section that decodes parameters representing spectrum characteristics coded by a speech coder;

a fixed excitation codebook;

an adaptive excitation codebook that represents a cyclic component of a synthesized speech signal; and

a section that decodes excitation vectors determined by said speech coder from said fixed excitation codebook and said adaptive excitation codebook and synthesizes a synthesized speech signal from the decoded excitation vector and said parameter.

15. An LPC parameter vector quantization apparatus comprising:

a distortion calculation section that calculates a Euclidean distortion between codevectors successively read from an LSF codebook storing many LSF codevectors and a quantization target;

an LPC code pre-selection section that preliminarily selects codevector indices that reduce said Euclidean distortion and generates decoded LSF parameters corresponding to their respective preliminarily selected indices;

a first LSF/LPC coefficient conversion section that converts said decoded LSF parameters to their respective decoded LPC coefficients;

a first LPC coefficient/LPC cepstrum conversion section that converts said decoded LPC coefficients to their respective LPC cepstra;

a second LSF/LPC coefficient conversion section that converts an LSF parameter which is a quantization target to an LPC coefficient;

a second LPC coefficient/LPC cepstrum conversion section that converts the LPC coefficient obtained by converting said quantization target to an LPC cepstrum;

an LPC code final search section that calculates cepstrum distortion (CD) between LPC cepstra obtained by converting the quantization targets supplied from said first and second LPC coefficient/LPC cepstrum conversion sections and the LPC cepstra obtained by converting the decoded LSF parameters, and outputs the index of the LSF codevector corresponding to the LPC cepstrum that minimizes said cepstrum distortion as an LPC code.

17. An LPC parameter decoder comprising:

an input section that inputs an LPC code output by the LSF parameter vector quantization apparatus according to claim 15;

and

a reading section that reads an arbitrary codevector from the same codebook as the codebook used for vector quantization and a plurality of codevectors stored in said codebook,

wherein the codevector corresponding to said input LPC code is read from said codebook and output as a decoded LPC parameter.

16. An LPC parameter vector quantization apparatus comprising:

a distortion calculation section that calculates a Euclidean distortion between codevectors successively read from an LSF codebook storing many LSF codevectors and a quantization target;

an LPC code pre-search section that preliminarily selects codevector indices that reduce said Euclidean distortion and generates decoded LSF parameters corresponding to their respective preliminarily selected indices;

an LSF/LPC coefficient conversion section that converts said decoded LSF parameters to their respective decoded LPC coefficients;

a first LPC coefficient/FFT power spectrum calculation section that calculates FFT power spectra from said decoded LPC coefficients;

an LSF/LPC coefficient conversion section that converts the LSF parameter which is a quantization target to an LPC coefficient;

a second LPC coefficient/FFT power spectrum calculation section that calculates an FFT power spectrum from the LPC coefficient obtained by converting said quantization target; and

an LPC code final search section that calculates spectrum distortion (SD) between FFT power spectra obtained by converting the quantization targets supplied from said first and second LPC coefficient/FFT power spectrum calculation sections and the FFT power spectra obtained by converting the decoded LSF parameters, and outputs the LSF codevector index according to the FFT power spectrum that minimizes said spectrum distortion as an LPC code.

22. An LSF parameter vector quantization apparatus having a 2-stage split configuration that calculates a differential vector by differentiating a prestored average LSF parameter from an input LSF parameter, calculates a predictive vector corresponding to said differential vector, calculates a predictive residual vector by further differentiating said predictive vector from said differential vector, vector-quantizes the predictive residual vector and outputs an LPC code, comprising:

a first stage vector quantization section that vector-quantizes said predictive residual vector using a plurality of predictive residual vectors stored in a first stage codebook and using a weighted Euclidean distortion as a distortion evaluation measure; and

a second stage vector quantization section that vector-quantizes a first stage VQ residual vector generated by said first stage vector quantization section using a plurality of first stage VQ residual vector lower frequency components stored in a second stage lower frequency codebook and first stage VQ residual vector higher frequency components stored in a second stage higher frequency codebook and using cepstrum distortion (CD) as a distortion evaluation measure.

25. An LSF parameter decoder that decomposes the LPC code output by the vector quantization apparatus according to claim 22 into index information, calculates a decoded predictive residual vector based on this index information and generates a decoded LSF parameter based on this decoded predictive residual vector, wherein said LSF parameter decoder reads a first stage codevector corresponding to said index information, second stage lower frequency codevector and second stage higher frequency codevector from their respective codebooks, adds up said three codevectors read and thereby generates a decoded predictive residual vector.

27. An LPC coefficient decoder comprising a function of converting the decoded LSF parameter generated by the LSF parameter decoder according to claim 25 to a decoded LPC coefficient.

23. An LSF parameter vector quantization apparatus having a 2-stage split configuration that calculates a differential vector by differentiating a prestored average LSF parameter from an input LSF parameter, calculates a predictive vector corresponding to said differential vector, calculates a predictive residual vector by further differentiating said predictive vector from said differential vector, vector-quantizes the predictive residual vector and outputs an LPC code, comprising:

a first stage vector quantization section that vector-quantizes said predictive residual vector using a plurality of predictive residual vectors stored in a first stage codebook and using a weighted Euclidean distortion as a distortion evaluation measure; and

a second stage vector quantization section that vector-quantizes a first stage VQ residual vector generated by said first stage vector quantization section using the same number of entries of lower frequency scaling factors as predictive residual vectors stored in a lower frequency scaling factor codebook, the same number of entries of higher frequency scaling factors as predictive residual vectors stored in the first stage codebook stored in a higher frequency scaling factor codebook and higher frequency codevectors stored in the second stage higher frequency codebook and using cepstrum distortion (CD) as a distortion evaluation measure.

26. An LSF parameter decoder that decomposes the LPC code output by the vector quantization apparatus according to claim 23 into index information, calculates a decoded predictive residual vector based on this index information and generates a decoded LSF parameter based on this decoded predictive residual vector, wherein said LSF parameter decoder reads a first stage codevector corresponding to said index information, second stage lower frequency codevector, second stage higher frequency codevector and lower frequency scaling factor and high scaling factor from their respective codebooks, calculates a sum of products of said three codevectors read and two scaling factors and thereby generates a decoded predictive residual vector.

28. An LPC coefficient decoder that converts the decoded LSF parameter generated by the LSF parameter decoder according to claim 26 to a decoded LPC coefficient.

30. A computer-readable recording medium that stores an LPC parameter vector decoding program, said LPC parameter vector decoding program comprising:

a step of reading codevectors corresponding to the index information output by the vector quantization program according to claim 23 from a codebook; and

a step of outputting the read codevector as a decoded LPC parameter.

29. A computer-readable recording medium that stores an LPC parameter vector quantization program, said LPC parameter vector quantization program comprising:

a step of calculating distortion between a plurality of codevectors read successively from a codebook storing many codevectors and a quantization target using a first distortion evaluation measure;

a step of preliminarily selecting a codevector that reduces an amount of distortion by said first distortion evaluation measure from among said many codevectors;

a step of calculating distortion between the preliminarily selected codevectors and said quantization target using a second distortion evaluation measure which is different from said first distortion evaluation measure; and

a step of specifying a codevector that minimizes the amount of distortion by said second distortion evaluation measure from among the preliminarily selected codevectors and outputting index information of the specified codevector.