Robust vector quantization of line spectral frequencies

• US 5,651,026 A
• Filed: 06/27/1995
• Issued: 07/22/1997
• Est. Priority Date: 06/01/1992
• Status: Expired due to Term

• Current Assignee: Hughes Network Systems LLC (Echostar Corporation)
• Original Assignee: Hughes Electronics Corporation (Rtx Corporation)
• Inventors: Lin, Daniel, Swaminathan, Kumar
• Primary Examiner(s): Chin, Stephen
• Assistant Examiner(s): PHAN, HAI
• Application Number:
  US08/495,148
• Time in Patent Office: 756 Days
• Field of Search: 375/240, 375/241, 375/242, 375/243, 375/244, 375/250, 375/245, 375/254, 381/29-31, 381/36, 381/37, 381/39, 395/2.31, 395/2.32, 348/422, 348/384
• US Class Current: 375/240
• CPC Class Codes: G06Q 10/0875 Itemisation or classification of parts, supplies or services, e.g. bill of materials
• CPC Class Codes: G10L 19/07 Line spectrum pair [LSP] vocoders
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum

Abstract

A line spectral frequency (LSF) vector quantizer, having particular application in digital cellular networks (DCN), is provided for code excited linear predictive (CELP) speech encoders. The LSF vector quantizer is efficient in terms of bits employed, robust and effective in terms of performance across speakers and handsets, moderate in terms of complexity, and accommodates effective and simple built-in transmission error detection schemes. The LSF vector quantizer employs a minimum number of bits, is of moderate complexity and incorporates built-in error detection capability in order to combat transmission errors. The LSF vector quantizer classifies unquantized line spectral frequencies into four categories, employing different vector quantization tables for each category. Each quantization table is optimized for particular types of vectors. For each category, three split vector codebooks are used with a simplified error measure to find three candidate split quantized vectors. The three sets of three split vectors are combined to produce as many as 27 vectors from each category. The quantizer then makes a final selection of optimal category using a more complex error measure to achieve the robust performance across speakers and handsets. Split vector quantization follows a two stage constrained search procedure that results in an ordered set of quantized line spectral frequencies that is "close" to the unquantized set with moderate complexity within each category. Effective and simple transmission error detection schemes at the receiver are made possible by the split nature of the vector quantization and the constrained search procedure. Only twenty-six bits are required to encode ten line spectral frequencies.

Claims

1. A method for quantizing line spectral frequency (LSF) vectors in a digital communications system comprising:

receiving an unquantized LSF vector;

searching each of a plurality of specialized codebooks and selecting from each of the plurality of specialized codebooks, a candidate quantized LSF vector resembling the unquantized LSF vector;

calculating a distortion measure for each of the candidate quantized LSF vectors using the unquantized LSF vector; and

selecting a quantized LSF vector from among the candidate quantized LSF vectors using the calculated distortion measures for each candidate quantized LSF vector.

2. The method of claim 1, wherein each specialized codebook comprises a plurality of specialized split codebooks, and wherein the step of searching each of the plurality of specialized codebooks comprises:

splitting the unquantized LSF vector into a plurality of unquantized split LSF vectors;

for each specialized codebook, searching each of the corresponding plurality of specialized split codebooks and selecting candidate quantized split LSF vectors resembling the corresponding unquantized split LSF vectors; and

combining the candidate quantized split LSF vectors to produce the candidate quantized LSF vectors.

3. The method of claim 2 wherein the step of searching each of the plurality of specialized codebooks comprises, after the step of searching each of the plurality of specialized split codebooks:

calculating a second distortion measure for each candidate quantized split LSF vector using the corresponding unquantized split LSF vector;

selecting a subset of candidate quantized split LSF vectors from among the candidate quantized split LSF vectors using the second distortion measure for each candidate split LSF vector; and

wherein the step of combining the candidate split LSF vectors comprises combining candidate split LSF vectors from among the subset of candidate quantized split LSF vectors so as to produce a plurality of unsplit candidate quantized LSF vectors.

4. The method of claim 3, wherein the step of combining the candidate split LSF vectors comprises combining split LSF vectors obtained from a split codebook corresponding to a first type of specialized codebook only with candidate split LSF vectors obtained from other split codebooks corresponding to the same first type.

5. The method of claim 3 wherein the second distortion measure calculated for a candidate quantized split LSF vector comprises a mean square calculation between the candidate quantized split LSF vector and the corresponding unquantized split LSF vector.

6. The method of claim 3 wherein the second distortion measure comprises a less accurate distortion measure than the first distortion measure.

7. The method of claim 1 wherein the distortion measure calculated for a candidate quantized LSF vector comprises a cepstral distance measurement between the candidate quantized LSF vector and the unquantized LSF vector.

8. The method of claim 1 wherein the specialized codebooks comprise a first codebook specialized for voiced speech and a second codebook specialized for unvoiced speech.

9. The method of claim 1 wherein the specialized codebooks comprise a first codebook specialized for filtered speech and a second codebook specialized for unfiltered speech.

10. The method of claim 9 wherein the first codebook is specialized for modeling the high pass filtering effects on speech of a telephone handset transducer.

11. The method of claim 2 wherein the plurality of unquantized split LSF vectors comprises three split unquantized split LSF vectors and wherein each plurality of specialized split codebooks comprises three specialized split codebooks for each specialized codebook.

12. A line spectral frequency (LSF) vector quantizer comprising:

an input for receiving an unquantized LSF vector;

a plurality of specialized codebooks of quantized (line spectral frequency) LSF vectors;

means coupled to the input for searching each of the specialized codebooks and selecting from each of the specialized codebooks, a candidate quantized LSF vector resembling the input unquantized (line spectral frequency) LSF vector;

means for calculating a distortion measure for each of the candidate quantized LSF vectors using the unquantized LSF vector; and

means for selecting a quantized LSF vector from among the candidate quantized LSF vectors using the calculated distortion measures for each candidate quantized LSF vector.

13. The quantizer of claim 12 wherein each specialized codebook comprises a plurality of specialized split codebooks of candidate quantized split LSF vectors, and wherein the means for searching each of the plurality of specialized codebooks comprises:

means for splitting the unquantized LSF vector into a plurality of unquantized split LSF vectors;

for each specialized codebook, means for searching each of the corresponding plurality of specialized split codebooks and selecting candidate quantized split LSF vectors resembling the corresponding unquantized split LSF vectors; and

means for combining the candidate quantized split LSF vectors to produce the candidate quantized LSF vectors.

14. The quantizer of claim 13 wherein the means for searching each of the plurality of specialized codebooks comprises:

means for calculating a second distortion measure for each candidate quantized split LSF vector using the corresponding unquantized split LSF vector;

means for selecting a subset of candidate quantized split LSF vectors from among the candidate quantized split LSF vectors using the second distortion measure for each candidate split LSF vector; and

wherein the means for combining the candidate split LSF vectors comprises means for combining candidate split LSF vectors chosen from among the subset of candidate quantized split LSF vectors so as to produce a plurality of unsplit candidate quantized LSF vectors.

15. The quantizer of claim 14, wherein the means for combining the candidate split LSF vectors comprises means for combining candidate split LSF vectors obtained from a split codebook corresponding to a first type of specialized codebook only with candidate split LSF vectors obtained from other split codebooks corresponding to the same first type.

16. The quantizer of claim 14 wherein the second distortion measure calculated for a candidate quantized split LSF vector comprises a mean square calculation between the candidate quantized split LSF vector and the corresponding unquantized split LSF vector.

17. The quantizer of claim 16 wherein the second distortion measure comprises a less accurate distortion measure than the first distortion measure.

18. The quantizer of claim 12 wherein the distortion measure calculated for a candidate quantized LSF vector comprises a cepstral distance measurement between the candidate quantized LSF vector and the unquantized LSF vector.

19. The quantizer of claim 12 wherein the specialized codebooks comprise a first codebook specialized for voice speech and a second codebook specialized for unvoiced speech.

20. The quantizer of claim 12 wherein the specialized codebooks comprise a first codebook specialized for filtered speech and a second codebook specialized for unfiltered speech.

21. A method for quantizing line spectral frequency (LSF) vectors in a digital communications system comprising:

receiving an unquantized LSF vector;

splitting the unquantized LSF vector into a plurality of unquantized split LSF vectors;

searching a plurality of split codebooks for candidate quantized split LSF vectors resembling each of the unquantized split LSF vectors;

calculating a first distortion measure for each candidate quantized split LSF vector using the corresponding unquantized split LSF vector;

selecting a subset of candidate quantized split LSF vectors from among the candidate quantized split LSF vectors based on the first distortion measure calculated for each candidate quantized split LSF vector;

combining candidate quantized split LSF vectors chosen from among the subset of candidate quantized split LSF vectors so as to produce a plurality of unsplit candidate quantized LSF vectors;

calculating a second distortion measure for each of the unsplit candidate quantized LSF vectors using the unquantized LSF vector; and

selecting a quantized LSF vector from among the unsplit candidate quantized LSF vectors using the second distortion measure calculated for each unsplit candidate quantized LSF vector.

22. The method of claim 21 wherein a specialized codebook comprises a plurality of split codebooks, and the step of searching each of the plurality of split codebooks comprises searching each of a plurality of specialized codebooks.

23. The method of claim 22 wherein the step of combining the candidate quantized split LSF vectors comprises combining candidate quantized split LSF vectors obtained from a split codebook corresponding to a first type of specialized codebook only with candidate quantized split LSF vectors obtained from other split codebooks corresponding to the same first type.

24. The method of claim 21, wherein the first distortion measure comprises a mean square calculation between each candidate quantized split LSF vector and the corresponding unquantized split LSF vector.

25. The method of claim 21 wherein the first distortion measure comprises a less accurate distortion measure than the second distortion measure.

26. The method of claim 21 wherein the second distortion measure comprises a cepstral distance measurement between each unsplit candidate quantized LSF vector and the unquantized LSF vector.

27. The method of claim 22 wherein the specialized codebooks comprise a first codebook specialized for voiced speech and a second codebook specialized for unvoiced speech.

28. The method of claim 22 wherein the specialized codebooks comprise a first codebook specialized for filtered speech and a second codebook specialized for unfiltered speech.

29. The method of claim 28 wherein the first codebook is specialized for modeling high pass filtering effects of a telephone handset transducer on speech.

30. The method of claim 22 wherein the plurality of split LSF vectors comprises three split LSF vectors and wherein each plurality of specialized split codebooks comprises three split codebooks for each specialized codebook.

31. A line spectral frequency (LSF) quantizer comprising:

a plurality of split codebooks of quantized split LSF vectors;

an input for receiving an unquantized LSF vector;

means for splitting the unquantized LSF vector into a plurality of unquantized split LSF vectors;

means coupled to the input for searching the plurality of split codebooks for candidate quantized split LSF vectors resembling each of the unquantized split LSF vectors;

means for calculating a first distortion measure for each candidate quantized split LSF vector using the corresponding unquantized split LSF vector;

means for selecting a subset of candidate quantized split LSF vectors from among the candidate quantized split LSF vectors based on the first distortion measure calculated for each candidate quantized split LSF vector;

means for combining candidate quantized split LSF vectors chosen from among the subset of candidate quantized split LSF vectors so as to produce a plurality of unsplit candidate quantized LSF vectors;

means for calculating a second distortion measure for each of the unsplit candidate quantized LSF vectors using the unquantized LSF vector; and

means for selecting a quantized LSF vector from among the unsplit candidate quantized LSF vectors using the second distortion measure calculated for each unsplit candidate quantized LSF vector.

32. The quantizer of claim 31 wherein a specialized codebook comprises a plurality of split codebooks, and means for searching the plurality of split codebooks comprises means for searching each of a plurality of specialized codebooks.

33. The quantizer of claim 32 wherein the means for combining the candidate quantized split LSF vectors combines candidate quantized split LSF vectors obtained from a split codebook corresponding to a first type of specialized codebook only with candidate quantized split LSF vectors obtained from other split codebooks corresponding to the same first type.

34. The quantizer of claim 37 wherein the first distortion measure comprises a mean square calculation between each candidate quantized split LSF vector and the corresponding unquantized split LSF vector.

35. The quantizer of claim 31 wherein the first distortion measure comprises a less accurate distortion measure than the second distortion measure.

36. The quantizer of claim 31 wherein the second distortion measure comprises a cepstral distance measurement between each unsplit candidate quantized LSF vector and the unquantized LSF vector.

37. The quantizer of claim 32 wherein the specialized codebooks comprise a first codebook specialized for voiced speech and a second codebook specialized for unvoiced speech.

38. The quantizer of claim 32 wherein the specialized codebooks comprise a first codebook specialized for filtered speech and a second codebook specialized for unfiltered speech.

39. A method for quantizing line spectral frequency (LSF) vectors in a digital communications system comprising:

receiving an unquantized LSF vector;

splitting the unquantized LSF vector into first, second and third unquantized split LSF vectors;

searching a plurality of split codebooks for candidate quantized split LSF vectors most closely resembling each of the first, second and third unquantized split LSF vectors;

calculating a first distortion measure for each of the first, second and third candidate quantized split LSF vectors using the corresponding unquantized split LSF vector;

selecting a first, second and third quantized split LSF vectors from among the candidate quantized split LSF vectors based on the first distortion measure calculated for each candidate split LSF vector;

combining candidate quantized split LSF vectors chosen from among the first, second and third quantized split LSF vectors so as to produce a plurality of unsplit candidate quantized LSF vectors;

calculating a cepstral distortion measure for each of the unsplit candidate quantized LSF vectors using the unquantized LSF vector; and

selecting a quantized LSF vector from among the unsplit candidate quantized LSF vectors using the cepstral distortion measure calculated for each unsplit candidate quantized LSF vector.

40. The method of claim 39 wherein the step of combining candidate quantized split vectors is performed such that at least one of the combinations of candidates is an ordered set.

41. The method of claim 39 wherein the plurality of codebooks includes four split codebooks.

42. The method of claim 41 wherein:

a first of said four split codebooks comprises filtered voiced LSF vectors;

a second of said four split codebooks comprises filtered unvoiced LSF vectors;

a third of said four split codebooks comprises unfiltered voiced LSF vectors; and

a fourth of said four split codebooks comprises unfiltered unvoiced LSF vectors.

43. A line spectral frequency (LSF) quantizer comprising:

a plurality of split codebooks of quantized split LSF vectors;

an input for receiving an unquantized LSF vector;

means for splitting the unquantized LSF vector into first, second and third unquantized split LSF vectors;

means coupled to the input for searching the plurality of split codebooks for candidate quantized split LSF vectors most closely resembling each of the first, second and third unquantized split LSF vectors;

means for calculating a first distortion measure for each of the first, second and third candidate quantized split LSF vectors using the corresponding unquantized split LSF vector;

means for selecting a first, second and third quantized split LSF vectors from among the candidate quantized split LSF vectors based an the first distortion measure calculated for each candidate split LSF vector;

means for combining candidate quantized split LSF vectors chosen from among the first, second and third quantized split LSF vectors so as to produce a plurality of unsplit candidate quantized LSF vectors;

means for calculating a cepstral distortion measure for each of the unsplit candidate quantized LSF vectors using the unquantized LSF vector; and

means for selecting a quantized LSF vector from among the unsplit candidate quantized LSF vectors using the cepstral distortion measure calculated for each unsplit candidate quantized LSF vector.

44. The LSF quantizer of claim 43 wherein the means for combining candidate quantized split vectors combines at least one of the combinations of candidates as an ordered set.

45. The LSF quantizer of claim 43 wherein the plurality of codebooks includes four split codebooks.

46. The LSF quantizer of claim 43 wherein:

a first of said four split codebooks comprises filtered voiced LSF vectors;

a second of said four split codebooks comprises filtered unvoiced LSF vectors;

a third of said four split codebooks comprises unfiltered voiced LSF vectors; and

a fourth of said four split codebooks comprises unfiltered unvoiced LSF vectors.