Codebook tables for encoding and decoding

• US 6,574,593 B1
• Filed: 09/15/2000
• Issued: 06/03/2003
• Est. Priority Date: 09/22/1999
• Status: Expired due to Term

• Current Assignee: DigiMedia Tech, LLC (IP Investments Group LLC)
• Original Assignee: Conexant Systems Incorporated (Synaptics Incorporated)
• Inventors: Su, Huan-Yu, Shlomot, Eyal, Thyssen, Jes, Gao, Yang, Benyassine, Adil
• Primary Examiner(s): To, Doris H.
• Assistant Examiner(s): NOLAN, DANIEL A
• Application Number:
  US09/663,837
• Time in Patent Office: 991 Days
• Field of Search: 704/211, 704/219, 704/222, 704/503
• US Class Current: 704/222
• CPC Class Codes: G10L 19/00 Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis in musical instruments G10H
• CPC Class Codes: G10L 19/167 Audio streaming, i.e. formatting and decoding of an encoded audio signal representation into a data stream for transmission or storage purposes
• CPC Class Codes: G10L 19/24 Variable rate codecs, e.g. for generating different qualities using a scalable representation such as hierarchical encoding or layered encoding
• CPC Class Codes: H03G 3/00 Gain control in amplifiers or frequency changers

Abstract

A speech compression system capable of encoding a speech signal into a bitstream for subsequent decoding to generate synthesized speech is disclosed. The speech compression system optimizes the bandwidth consumed by the bitstream by balancing the desired average bit rate with the perceptual quality of the reconstructed speech. The speech compression system comprises a full-rate codec, a half-rate codec, a quarter-rate codec and an eighth-rate codec. The codecs are selectively activated based on a rate selection. In addition, the full and half-rate codecs are selectively activated based on a type classification. Each codec is selectively activated to encode and decode the speech signals at different bit rates emphasizing different aspects of the speech signal to enhance overall quality of the synthesized speech.

Claims

1. A variable rate speech compression system for processing a speech signal, the variable rate speech compression system comprising:

an encoding system operable to determine a rate selection and a type classification for the speech signal, the encoding system comprising;

a line spectrum frequency prediction error quantization table selectable as a function of the rate selection, the line spectrum frequency prediction error quantization table associated with encoding short-term predictor parameters of the speech signal;

a 2D gain quantization table associated with jointly encoding an adaptive codebook gain and a fixed codebook gain of the speech signal when the type classification is a first type;

a pre-gain quantization table selectable as a function of the rate selection, the pre-gain quantization table associated with exclusively encoding the adaptive codebook gain when the type classification is a second type;

a delayed gain quantization table selectable as a function of the rate selection, the delayed gain quantization table associated with exclusively encoding the fixed codebook gain when the type classification is the second type; and

a decoding system in communication with the encoding system, the decoding system operable to decode the speech signal with the line spectrum frequency prediction error quantization table and at least one of;

the 2D gain quantization table,

the pre-gain quantization table, and

the delayed gain quantization table,

as a function of the rate selection and the type classification.

2. The variable rate speech compression system of claim 1, where the line spectrum frequency prediction error quantization table comprises four stages when the rate selection is a full rate.

3. The variable rate speech compression system of claim 2, where a first stage comprises 128 quantization vectors.

4. The variable rate speech compression system of claim 1, where the line spectrum frequency prediction error quantization table comprises a first stage and a second stage each with 128 quantization vectors and a third stage with 64 quantization vectors when the rate selection is a half rate.

5. The variable rate speech compression system of claim 4, where the first stage comprises a first quantization vector represented as {0.00842379, 0.00868718, 0.01533677, 0.00423439, −

0.00886805, −

0.02132286, −

0.03152681, −

0.01975061, −

0.01152093, −

0.01341948} and a second quantization vector represented as {0.02528175, 0.04259634, 0.03789221, 0.01659535, −

0.00266498, −

0.01529545, −

0.01653101, −

0.01528401, −

0.01047642, −

0.01127117}.

6. The variable rate speech compression system of claim 4, where the second stage comprises a first quantization vector represented as {0.00589332, 0.00462334, −

0.00937151, −

0.01478366, 0.00674597, 0.00164302, −

0.00890749, −

0.00091839, 0.00487032, 0.00012026} and a second quantization vector represented as {−

0.00346857, −

0.00100200, −

0.00418711, −

0.01512477, −

0.00104209, −

0.00491133, −

0.00209555, 0.00045850, 0.00023339, 0.00567173}.

7. The variable rate speech compression system of claim 4, where the third stage comprises a first quantization vector represented as {−

0.00071405, 0.00244371, 0.00235739, −

0.00329369, 0.00472867, −

0.00361321, −

0.00584670, 0.00863128, 0.00145642, −

0.00441746} and a second quantization vector represented as {0.00242589, −

0.00430711, −

0.00122645, −

0.00464764, −

0.00017887, −

0.00471663, 0.00181162, 0.00249980, −

0.00276848, −

0.00485697}.

8. The variable rate speech compression system of claim 1, where the encoding system is operable to jointly encode the fixed codebook gain and the adaptive codebook gain with the 2D gain quantization table for each of at least two subframes of a frame of the speech signal.

9. The variable rate speech compression system of claim 1, where the 2D gain quantization table comprises 128 quantization vectors of 2 elements each.

10. The variable rate speech compression system of claim 1, where the 2D gain quantization table comprises a first quantization vector represented as {1.13718400, 2.00167200} and a second quantization vector represented as {1.15061100, 0.80219900} when the rate selection is a full rate.

11. The variable rate speech compression system of claim 1, where the pre-gain quantization table comprises 64 vectors when the rate selection is a full rate.

12. The variable rate speech compression system of claim 1, where the pre-gain quantization table comprises 16 vectors when the rate selection is a half rate.

13. The variable rate speech compression system of claim 1, where the pre-gain quantization table comprises a first vector represented as {0.60699869, 0.59090763, 0.64920781, 0.64610492} and a second vector represented as {0.68101613, 0.65403889, 0.64210982, 0.63130892} when the rate selection is a full rate.

14. The variable rate speech compression system of claim 1, where the pre-gain quantization table comprises a first vector represented as {1.16184904, 1.16859789, 1.13656320} and a second vector represented as {1.14613289, 1.06371877, 0.91852525} when the rate selection is a half rate.

15. The variable rate speech compression system of claim 1, where the delayed gain quantization table comprises 1024 vectors when the rate selection is a full rate.

16. The variable rate speech compression system of claim 1, where the delayed gain quantization table comprises 256 vectors when the rate selection is a half rate.

17. The variable rate speech compression system of claim 1, where the encoding system is operable to encode the frame with the delayed gain quantization table and a plurality of predictor coefficients when the type classification is the second type.

18. The variable rate speech compression system of claim 17, where the predictor coefficients comprise a first predictor coefficient represented as {0.7, 0.6, 0.4, 0.2}, a second predictor coefficient represented as {0.4, 0.2, 0.1, 0.05}, a third predictor coefficient represented as {0.3, 0.2, 0.075, 0.025} and a fourth predictor coefficient represented as {0.2, 0.075, 0.025, 0.0} when the rate selection is a full rate.

19. The variable rate speech compression system of claim 17, where the predictor coefficients comprise a first predictor coefficient represented as {0.6, 0.3, 0.1}, a second predictor coefficient represented as {0.4, 0.25, 0.1}, and a third predictor coefficient represented as {0.3, 0.15, 0.075} when the rate selection is a half rate.

20. The variable rate speech compression system of claim 1, where the delayed gain quantization table comprises a first vector represented as {0.18423671, 0.06523999, 0.13390472} and a second vector represented as {0.27552690, 0.09702324, 0.05427950} when the rate selection is a half rate.

21. The variable rate speech compression system of claim 1, where the encoding system further comprises a line spectrum frequency predictor coefficients table associated with encoding short-term predictor parameters, the line spectrum frequency predictor coefficients table comprising:

a first set of predictor coefficients; and

a second set of predictor coefficients;

wherein the first and the second set of predictor coefficients are selectable by the encoding system when the rate selection is a half rate.

22. The variable rate speech compression system of claim 21, where the first set of predictor coefficients comprises a first vector represented as {0.45782564, 0.59002827, 0.73704688, 0.73388197, 0.75903791, 0.74076479, 0.65966007, 0.58070788, 0.52280647, 0.42738207} and a second vector represented as {0.19087084, 0.26721569, 0.38110463, 0.39655069, 0.43984539, 0.42178869, 0.34869783, 0.28691864, 0.23847475, 0.17468375}.

23. The variable rate speech compression system of claim 21, where the second set of predictor coefficients comprises a first vector represented as {0.14936742, 0.25397094, 0.42536339, 0.40318214, 0.39778242, 0.34731435, 0.22773174, 0.17583478, 0.12497067, 0.11001108} and a second vector represented as {0.09932127, 0.15389237, 0.24021347, 0.24507006, 0.26478926, 0.23018456, 0.15178193, 0.11368182, 0.07674584, 0.06122567}.

24. A variable rate speech compression system for processing a speech signal, the variable rate speech compression system comprising:

an encoding system operable to determine a bit rate and a type classification for the speech signal, the bit rate comprising a first rate and a second rate, and the type classification comprising a first type and a second type, the encoding system comprising;

a line spectrum frequency prediction error quantization table selectable as a function of the bit rate, wherein the encoding system is operable to encode short-term predictor parameters of the speech signal with the line spectrum frequency prediction error quantization table;

an interpolation module operable with the line spectrum frequency prediction error quantization table to encode short-term predictor parameters, when the bit rate is the first rate and the type classification is the first type;

a line spectrum frequency predictor coefficient table selectable as a function of the bit rate, wherein the encoding system is operable to generate predicted line spectrum frequencies with the line spectrum frequency predictor coefficient table; and

a predictor switch module operable with the line spectrum frequency predictor coefficient table to generate predicted line spectrum frequencies, when the bit rate is the second rate.

25. The variable rate speech compression system of claim 24, where the line spectrum frequency prediction error quantization table comprises 4 stages when the bit rate is the first rate.

26. The variable rate speech compression system of claim 24, where the line spectrum frequency prediction error quantization table comprises 3 stages when the bit rate is the second rate.

27. The variable rate speech compression system of claim 24, where the line spectrum frequency predictor coefficients table comprises two vectors of predictor coefficients when the bit rate is the first rate.

28. The variable rate speech compression system of claim 24, where the line spectrum frequency predictor coefficients table comprises two sets of four vectors of predictor coefficients when the bit rate is the second rate, where the predictor switch is operable to select one of the sets.

29. The variable rate speech compression system of claim 24, where the interpolation module comprises a plurality of interpolation paths selectable as a function of variations in the spectral envelope of the speech signal.

30. The variable rate speech compression system of claim 29, where the interpolation paths comprise 4 interpolation paths.

31. The variable rate speech compression system of claim 24, where the interpolation module is operable to apply one of a plurality of interpolation paths to adjust the contour of a spectral envelope of the speech signal.

32. The variable rate speech compression system of claim 31, where the interpolation module is operable to determine the interpolation paths as a function of a plurality of predetermined weighting factors.

33. A method of processing a speech signal with a variable rate speech compression system, the method comprising:

determining a rate and a type for the speech signal;

encoding short-term predictor parameters of the speech signal with a line spectrum frequency prediction error quantization table as a function of the rate;

jointly encoding an adaptive codebook gain and a fixed codebook gain of the speech signal with a 2D gain quantization table when the type is a first type;

encoding the adaptive codebook gain with a pre-gain quantization table as a function of the rate when the type is a second type; and

encoding the fixed codebook gain with a delayed gain quantization table as a function of the rate when the type is the second type.

34. The method of claim 33, further comprising decoding the speech signal with the line spectrum frequency prediction error quantization table and at least one of the 2D gain quantization table, the pre-gain quantization table and the delayed gain quantization table as a function of the rate and the type.

35. The method of claim 33, where encoding with the line spectrum frequency prediction error quantization table when the rate is a half rate comprises:

selecting one of a first set of predictor coefficients and a second set of predictor coefficients from a line spectrum frequency predictor coefficients table; and

determining predicted line spectrum frequencies with the selected set of predictor coefficients.

36. The method of claim 33, where encoding with the line spectrum frequency prediction error quantization table when the rate is a full rate and the type is the first type comprises:

selecting one of a plurality of interpolation paths; and

adjusting the weighting of previously quantized line spectrum frequencies and the weighting of currently quantized line spectrum frequencies with the interpolation path.

37. The method of claim 33, where encoding with the pre-gain quantization table when the rate is a full rate comprises:

determining the adaptive codebook gain for each of four subframes of a frame of the speech signal; and

analyzing vectors in the pre-gain quantization table comprising a first vector represented as {0.60699869, 0.59090763, 0.64920781, 0.64610492} to select one of the vectors with elements representing the adaptive codebook gain of each of the subframes.

38. The method of claim 33, where encoding with the pre-gain quantization table when the rate is a half rate comprises:

determining the adaptive codebook gain for each of three subframes of a frame of the speech signal; and

analyzing vectors in the pre-gain quantization table comprising a first vector represented as {1.16184904, 1.16859789, 1.13656320} to select one of the vectors with elements representing the adaptive codebook gain of each of the subframes.

39. The method of claim 33, where encoding with the delayed gain quantization table when the rate is a half rate comprises:

completing the search in a fixed codebook for each of three subframes of a frame of the speech signal;

determining the fixed codebook gain for each of the subframes; and

analyzing vectors in the delayed gain quantization table comprising a first vector represented as {0.18423671, 0.06523999, 0.13390472} to select one of the vectors with elements representing the fixed codebook gain of each of the subframes.

40. The method of claim 33, here encoding with the delayed gain quantization table when the type is the second type comprising:

representing the fixed codebook gain for each of a plurality of subframes of a frame of the speech signal with a fixed codebook energy;

generating a predicted fixed codebook energy for each of the subframes with quantized fixed codebook energy errors from a plurality of subframes of a previous frame and a plurality of predictor coefficients;

forming a vector with the difference in the fixed codebook energy and the predicted fixed codebook energy; and

selecting a corresponding vector from the delayed gain quantization table.

41. The method of claim 40, where generating the predicted fixed codebook energy comprises multiplying the quantized fixed codebook energy errors by the predictor coefficients, the predictor coefficients comprising a first subframe predictor coefficient represented as {0.7, 0.6, 0.4, 0.2}, a second subframe predictor coefficient represented as {0.4, 0.2, 0.1, 0.05}, a third subframe predictor coefficient represented as {0.3, 0.2, 0.075, 0.025} and a fourth subframe predictor coefficient represented as {0.2, 0.075, 0.025, 0.0}.

42. The method of claim 40, where generating the predicted fixed codebook energy comprises multiplying the quantized fixed codebook energy errors by the predictor coefficients, the predictor coefficients comprising:

a first predictor coefficient represented as {0.6, 0.3, 0.1};

a second predictor coefficient represented as {0.4, 0.25, 0.1}; and

a third predictor coefficient represented as {0.3, 0.15, 0.075};

wherein the rate selection is a half rate.

43. The method of claim 33, where jointly encoding with the 2D gain quantization table when the rate is the full rate comprises analyzing vectors within the 2D gain quantization table, the vectors comprising a first vector represented as {1.13718400, 2.00167200}.

44. A method of processing a speech signal, the method comprising:

selecting a bit rate and a type classification;

converting short-term predictor parameters extracted from the speech signal to line spectrum frequencies;

determining predicted line spectrum frequencies with a line spectrum frequency predictor coefficients table when the bit rate selected is a first rate;

determining predicted line spectrum frequencies with a line spectrum frequency predictor coefficients table and a predictor switch module when the bit rate selected is a second rate;

subtracting predicted line spectrum frequencies from line spectrum frequencies to generate a line spectrum frequencies prediction error;

quantizing the line spectrum frequencies predication error to produce quantized line spectrum frequencies; and

modifying the quantized line spectrum frequencies with an interpolation module when the bit rate selected is the first rate and the type classification is a first type;

wherein when the bit rate is the second rate, determining predicted line spectrum frequencies comprises selecting one of;

a first set of predictor coefficients, the first set of predictor coefficients including a first vector represented as {0.45782564, 0.59002827, 0.73704688, 0.73388197, 0.75903791, 0.74076479, 0.65966007, 0.58070788, 0.52280647, 0.42738207}; and

a second set of predictor coefficients, the second set of predictor coefficients including a first vector represented as {0.14936742, 0.25397094, 0.42536339, 0.40318214, 0.39778242, 0.34731435, 0.22773174, 0.17583478, 0.12497067, 0.11001108}.

45. The method of claim 44, where determining predicted line spectrum frequencies when the bit rate is the second rate comprises selecting a set of predictor coefficients from the line spectrum frequency predictor coefficients table with the predictor switch module.

46. The method of claim 44, where modifying the quantized line spectrum frequencies comprises selecting one of a plurality of interpolation paths, the interpolation paths derived from a predetermined weighting factor.

47. The method of claim 44, where modifying the quantized line spectrum frequencies comprises:

analyzing the degree of spectral variations between a plurality of subframes of a frame of the speech signal; and

selecting an interpolation path as a function of the spectral variations.