Adaptive-block-length, adaptive-transforn, and adaptive-window transform coder, decoder, and encoder/decoder for high-quality audio

• US 5,394,473 A
• Filed: 12/11/1991
• Issued: 02/28/1995
• Est. Priority Date: 04/12/1990
• Status: Expired due to Term

• Current Assignee: Dolby Laboratories Licensing Corporation (Dolby Laboratories Incorporated)
• Original Assignee: Dolby Laboratories Licensing Corporation (Dolby Laboratories Incorporated)
• Inventors: Davidson, Grant A.
• Primary Examiner(s): Richardson, Robert L.
• Assistant Examiner(s): Tung, Kee M.
• Application Number:
  US07/781,262
• Time in Patent Office: 1,175 Days
• Field of Search: 381/29-40, 381/41, 381/51, 341/51, 341/67, 375/122, 395/2.38, 395/2.67, 395/2.13
• US Class Current: 704/200.1
• CPC Class Codes: G10L 19/0212 using orthogonal transformation
• CPC Class Codes: G10L 19/022 Blocking, i.e. grouping of samples in time; Choice of analysis windows; Overlap factoring
• CPC Class Codes: H04B 1/66 for reducing bandwidth of signals; for improving efficiency of transmission H04B1/68 takes precedence

Abstract

The invention relates in general to high-quality low bit-rate digital transform coding and decoding of information corresponding to audio signals such as music signals. More particularly, the invention relates to signal analysis/synthesis in coding and decoding. The invention can optimize the trade off in transform coders between time resolution and frequency resolution by adaptively selecting the transform block length for each sampled audio segment, and/or can optimize coding gain by adaptively selecting the transform and/or by adaptively selecting the analysis window or the analysis/synthesis window pair.

Claims

1. An encoder for the encoding of samples representing a discrete time signal and particularly a music signal, comprising

control means responsive to one or more characteristics of said discrete time signal for adapting at least one of a sample block length, one or more analysis-window functions, and one or more discrete transforms, and for grouping said samples into sample blocks having a length equal to said sample block length, wherein said control means adapts said sample block length for any sample block such that it is different from the length of a previous sample block.

analysis-window means responsive to said control means for generating analysis-window weighted blocks by weighting said sample blocks with said one or more analysis-window functions, and

transform means responsive to said analysis-window means and optionally responsive to said control means for generating spectral information by applying said one or more discrete transforms to said analysis-window weighted blocks.

3. An encoder according to claim 1 or 2 wherein said control means comprises

means for segmenting each of said sample blocks into one or more levels of subblocks having different subblock lengths, and

means for detecting the occurrence of a transient whenever an amount of increase in amplitude or decrease in amplitude between a high-pass filtered representation of said discrete time signal in adjacent subblocks in each of said one or more levels of subblocks exceeds one or more trigger thresholds, and

wherein said analysis-window means weights subblocks with one or more analysis-window functions.

4. An encoder according to claim 3 wherein said one or more trigger thresholds comprise a hierarchy of thresholds, each level of said hierarchy of thresholds corresponding to a respective one of said one or more levels of subblocks.

5. An encoder according to claim 3 wherein said control means is responsive to the occurrence of said one or more characteristics only for portions of said discrete time signal having an amplitude which equals or exceeds a silence threshold.

6. An encoder according to claim 3 wherein said transform means effectively applies a single discrete transform to pairs of said analysis-window weighted blocks.

7. An encoder according to claim 6 wherein said control means adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more discrete transforms.

8. An encoder according to claim 1 or 2 wherein said control means comprises a means for detecting the occurrence of a transient whenever an amount of increase in amplitude of a high-pass filtered representation of said discrete time signal over an interval substantially the same as or less than said sample block length exceeds one or more attack thresholds, or an amount of decrease in amplitude of said high-pass filtered representation of said discrete time signal over an interval substantially the same as or less than said sample block length exceeds one or more decay thresholds.

9. An encoder according to claim 8 wherein said control means comprises a means for segmenting each of said sample blocks into one or more levels of subblocks having different subblock lengths, wherein said increase in amplitude and said decrease in amplitude are established from changes in amplitude between adjacent subblocks in each of said one or more levels of subblocks, and wherein said analysis-window means weights subblocks with one or more analysis-window functions.

10. An encoder according to claim 9 wherein said one or more attack thresholds and said one or more decay thresholds are a hierarchy of thresholds, each level of said hierarchy of thresholds corresponding to a respective one of said one or more levels of subblocks.

11. An encoder according to claim 9 wherein said control means is responsive to the occurrence of said one or more characteristics only for portions of said discrete time signal having an amplitude which equals or exceeds a silence threshold.

12. An encoder according to claim 9 wherein said transform means effectively applies a single discrete transform to pairs of said analysis-window weighted blocks.

13. An encoder according to claim 12 wherein said control means adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more discrete transforms.

14. An encoder according to claim 1 or 2 wherein said control means comprises

means for segmenting each of said sample blocks into subblocks, and

means for detecting in which of said subblocks an event having said one or more characteristics occurs.

15. An encoder according to claim 1 or 2 wherein said control means is responsive to the occurrence of said one or more characteristics only for portions of said discrete time signal having an amplitude which equals or exceeds a silence threshold.

16. An encoder according to claim 1 or 2 wherein said control means comprises means for detecting the occurrence of one or more events within said discrete time signal, said detecting in accordance with psychoacoustic principles.

17. An encoder according to claim 16 wherein said one or more events are abrupt amplitude changes in said discrete time signal.

18. An encoder according to claim 17 wherein said detecting is in accordance with pre- and post-event masking intervals of human hearing.

19. An encoder according to claim 1 or 2 wherein each of said one or more analysis-window functions has a respective shape, and wherein said control means adapts said one or more analysis-window functions by adapting one or more parameters affecting the respective shape of said one or more analysis-window functions and/or selecting said one or more analysis-window functions from a plurality of functions.

20. An encoder according to claim 1 or 2 wherein each of said discrete transforms has a respective kernel function, and wherein said control means adapts said one or more discrete transforms by adapting one or more parameters affecting the respective kernel function of said one or more discrete transforms and/or selecting said one or more discrete transforms from a plurality of transforms.

21. An encoder according to claim 1 or 2 wherein said control means comprises

means for establishing a frequency of a dominant spectral component of said discrete time signal,

means for determining a difference frequency between the frequency of said dominant spectral component and a center frequency of a subband for each transform in a set of transforms comprising a plurality of discrete transforms, and

means for selecting said one or more discrete transforms from said set of transforms such that said difference frequency is substantially minimized.

22. An encoder according to claim 1 or 2 wherein said analysis-window weighted blocks comprise real valued samples and the application of said one or more discrete transforms substantially corresponds to Evenly-Stacked Time-Domain Aliasing Cancellation, said transform means comprising

means for multiplying real valued samples in pairs of said analysis-window weighted blocks by a first complex function to obtain complex-valued modified sample sets,

means for applying a Fast Fourier Transform against said complex-valued modified sample sets, and

means for multiplying the results of the transformation by a second complex function to obtain said spectral information.

23. An encoder according to claim 1 or 2 wherein said analysis-window weighted blocks comprise real valued samples and the application of said one or more discrete transforms substantially corresponds to Oddly-Stacked Time-Domain Aliasing Cancellation, said transform means comprising

means for multiplying real valued samples in said analysis-window weighted blocks by a first complex function to obtain complex-valued modified sample sets,

means for applying a Fast Fourier Transform against said complex-valued modified sample sets, and

means for multiplying the results of the transformation by a second complex function to obtain said spectral information.

24. An encoder according to claim 1 or 2 wherein said transform means effectively applies a single discrete transform to pairs of said analysis-window weighted blocks.

25. An encoder according to claim 24 wherein said control means adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more discrete transforms.

26. An encoder according to claim 1 or 2 wherein said control means adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more discrete transforms.

2. An encoder for the encoding of samples representing a discrete time signal and particularly a music signal;

comprising

control means responsive to one or more characteristics of said discrete time signal for adapting a sample block length, for adapting one or more analysis-window functions and/or one or more discrete transforms, and for grouping said samples into sample blocks having a length equal to said sample block length,

analysis-window means responsive to said control means for generating analysis-window weighted blocks by weighting said sample blocks with said one or more analysis-window functions, and

transform means responsive to said analysis-window means and optionally responsive to said control means for generating spectral information by applying said one or more discrete transforms to said analysis-window weighted blocks.

27. A decoder for the decoding of a coded signal comprising representations of spectral components of a discrete time signal and particularly a music signal, wherein said decoder comprises

inverse control means responsive to said coded signal for adapting a spectral block length and one or more inverse discrete transforms, and for grouping said representations of spectral components into spectral blocks having a length equal to said spectral block length, and

transform means responsive to said inverse control means for generating sample blocks by applying said one or more inverse discrete transforms to said spectral blocks.

28. A decoder according to claim 27 wherein said inverse control means segments said spectral blocks into subblocks and said transform means applies one or more transforms to each of said subblocks.

29. A decoder according to claim 28 wherein said transform means effectively applies a single inverse discrete transform to pairs of said subblocks.

36. A decoder according to claim 27 or 30 wherein said transform means effectively applies a single inverse discrete transform to pairs of said spectral blocks.

30. A decoder for the decoding of a coded signal comprising representations of spectral components of a discrete time signal and particularly a music signal, wherein said decoder comprises

inverse control means responsive to said coded signal for adapting at least one of a spectral block length, one or more inverse discrete transforms, and one or more synthesis-window functions, and for grouping said representations of spectral components into spectral blocks having a length equal to said spectral block length, wherein said inverse control means adapts said spectral block length for any spectral block such that it is different from the length of a previous spectral block,

transform means responsive to said inverse control means for generating sample blocks comprising recovered signal samples by applying said one or more inverse discrete transforms to said spectral blocks, and

synthesis-window means responsive to said transform means and optionally responsive to said inverse control means for generating output samples by weighting said sample blocks with said one or more synthesis-window functions.

31. A decoder according to claim 30 wherein said inverse control means segments said spectral blocks into subblocks and said transform means applies one or more transforms to each of said subblocks.

32. A decoder according to claim 31 wherein said transform means effectively applies a single inverse discrete transform to pairs of said subblocks.

33. A decoder according to claim 30, 31, or 32 wherein said inverse control means adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more inverse discrete transforms.

34. A decoder according to claim 30 wherein said representations of spectral components are real valued and the application of said one or more inverse discrete transforms substantially corresponds to Evenly-Stacked Time-Domin Aliasing Cancellation, said transform means comprising

means for multiplying said representations of spectral components in pairs of said spectral blocks by a first complex function to obtain complex-valued modified spectral sets,

means for applying an Inverse Fast Fourier Transform against said complex-valued modified spectral sets, and

means for multiplying the results of the transformation by a second complex function to obtain pairs of said sample blocks.

35. A decoder according to claim 30 wherein said representations of spectral components are real valued and the application of said one or more inverse discrete transforms substantially corresponds to Oddly-Stacked Time-Domain Aliasing Cancellation, said transform means comprising

means for multiplying said representations of spectral components by a first complex function to obtain complex-valued modified spectral sets,

means for applying an Inverse Fast Fourier Transform against said complex-valued modified spectral sets, and

means for multiplying the results of the transformation by a second complex function to obtain said sample blocks.

37. An encoding method for the encoding of samples representing a discrete time signal and particularly a music signal, comprising

adapting at least one of a sample block length, one or more analysis-window functions, and one or more discrete transforms in response to one or more characteristics of said discrete time signal, and grouping said samples into sample blocks having a length equal to said sample block length, wherein said adapting adapts said sample block length for any sample block such that it is different from the length of a previous sample block,

generating analysis-window weighted blocks by weighting said sample blocks with said one or more analysis-window functions, and

generating spectral information by applying said one or more discrete transforms to said analysis-window weighted blocks.

39. An encoding method according to claim 37 wherein said adapting and grouping further comprises

segmenting each of said sample blocks into one or more levels of subblocks having different subblock lengths, and

detecting the occurrence of a transient whenever an amount of increase in amplitude or decrease in amplitude between a high-pass filtered representation of said discrete time signal in adjacent subblocks in each of said one or more levels of subblocks exceeds one or more trigger thresholds, and

wherein said generating analysis-window weighted blocks weights each subblock with one or more analysis-window functions.

41. An encoding method according to claim 39 or 40 wherein said adapting adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more discrete transforms, and wherein said generating spectral information effectively applies a single discrete transform to pairs of said analysis-window weighted blocks.

42. An encoding method according to claim 37 wherein said adapting and grouping further comprises segmenting each of said sample blocks into one or more levels of subblocks having different subblock lengths, wherein said adapting is in response to detecting the occurrence of a transient whenever an amount of increase in amplitude of a high-pass filtered representation of said discrete time signal between adjacent subblocks in each of said one or more levels of subblocks exceeds one or more attack thresholds, or an amount of decrease in amplitude of said high-pass filtered representation of said discrete time signal between adjacent subblocks in each of said one or more levels of subblocks exceeds one or more decay thresholds, and wherein said generating analysis-window weighted blocks weights subblocks with one or more analysis-window functions.

44. An encoding method according to claim 37 or 38 wherein said control means comprises detecting the occurrence of one or more abrupt amplitude changes in said discrete time signal, said detecting in accordance with psychoacoustic principles.

45. An encoding method according to claim 44 wherein said detecting is in accordance with pre- and post-event masking intervals of human hearing.

46. An encoding method according to claim 37 or 38 wherein said adapting comprises

establishing a frequency of a dominant spectral component of said discrete time signal,

determining a difference frequency between the frequency of said dominant spectral component and a center frequency of a subband for each transform in a set of transforms comprising a plurality of discrete transforms, and

selecting said one or more discrete transforms from said set of transforms such that said difference frequency is substantially minimized.

47. An encoding method according to claim 37 or 38 wherein said generating spectral information effectively applies a single discrete transform to pairs of said analysis-window weighted blocks.

48. An encoding method according to claim 37 or 38 wherein said adapting adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more discrete transforms.

38. An encoding method for the encoding of samples representing a discrete time signal and particularly a music signal, comprising

adapting a sample block length, adapting one or more analysis-window functions and/or one or more discrete transforms in response to one or more characteristics of said discrete time signal, and grouping said samples into sample blocks having a length equal to said sample block length,

generating analysis-window weighted blocks by weighting said sample blocks with said one or more analysis-window functions, and

generating spectral information by applying said one or more discrete transforms to said analysis-window weighted blocks.

40. An encoding method according to claim 38 wherein said segmenting comprises

segmenting each of said sample blocks into one or more levels of subblocks having different subblock lengths, and

detecting the occurrence of a transient whenever an amount of increase in amplitude or decrease in amplitude between a high-pass filtered representation of said discrete time signal in adjacent subblocks in each of said one or more levels of subblocks exceeds one or more trigger thresholds, and

wherein said generating analysis-window weighted blocks weights subblocks with one or more analysis-window functions.

43. An encoding method according to claim 38 wherein said segmenting segments each of said sample blocks into one or more levels of subblocks having different subblock lengths, wherein said adapting is in response to detecting the occurrence of a transient whenever an amount of increase in amplitude of a high-pass filtered representation of said discrete time signal between adjacent subblocks in each of said one or more levels of subblocks exceeds one or more attack thresholds, or an amount of decrease in amplitude of said high-pass filtered representation of said discrete time signal between adjacent subblocks in each of said one or more levels of subblocks exceeds one or more decay thresholds, and wherein said generating analysis-window weighted blocks weights subblocks with one or more analysis-window functions.

49. A decoding method for the decoding of a coded signal comprising representations of spectral components of a discrete time signal and particularly a music signal, wherein said decoding method comprises

adapting a spectral block length and one or more inverse discrete transforms in response to said coded signal, and grouping said representations of spectral components into spectral blocks having a length equal to said spectral block length, and

generating sample blocks by applying said one or more inverse discrete transforms to said spectral blocks.

51. A decoding method according to claim 49 wherein said adapting and grouping also segments said spectral blocks into subblocks, and wherein said generating sample blocks effectively applies a single inverse discrete transform to pairs of said subblocks.

54. A decoding method according to claim 49 or 50 wherein said generating sample blocks effectively applies a single inverse discrete transform to pairs of said spectral blocks.

50. A decoding method for the decoding of a coded signal comprising representations of spectral components of a discrete time signal and particularly a music signal, wherein said decoding method comprises

adapting in response to said coded signal at least one of a spectral block length, one or more inverse discrete transforms, and one or more synthesis-window functions in response to said coded signal, and grouping said representations of spectral components into spectral blocks having a length equal to said spectral block length, wherein said adapting adapts said spectral block length for any spectral block such that it is different from the length of a previous spectral block,

generating sample blocks comprising recovered signal samples by applying said one or more inverse discrete transforms to said spectral blocks, and

generating output samples by weighting said sample blocks with said one or more synthesis-window functions.

52. A decoding method according to claim 50 wherein said adapting and grouping also segments said spectral blocks into subblocks, and wherein said generating sample blocks effectively applies a single inverse discrete transform to pairs of said subblocks.

53. A decoding method according to claim 50 or 52 wherein said adapting adapts a Time-Domain Aliasing Cancellation phase parameter for at least one of said one or more inverse discrete transforms.

55. An encoder for the encoding of samples representing a discrete time signal and particularly a music signal, comprising

a signal characteristic detector responsive to said discrete time signal,

a frame control responsive to said signal characteristic detector comprising at least one of an analysis window function control and a discrete transform control,

a buffer responsive to said samples,

an analysis window responsive to said buffer and optionally responsive to said frame control, said analysis window comprising one or more analysis-window functions, and

a filter bank responsive to said analysis window and optionally responsive to said frame control, said filter bank comprising one or more discrete transforms,

wherein at least one of said analysis window and said filter bank is responsive to said frame control.

56. An encoder according to claim 55 wherein said signal characteristic detector comprises

a high-pass filter,

a hierarchical sample subblock segmenter responsive to said high-pass filter,

a peak-amplitude detector responsive to said hierarchical sample subblock segmenter, and

a threshold comparator responsive to said peak-amplitude detector.

57. An encoder according to claim 56 wherein said threshold comparator comprises

a silence-threshold comparator,

a peak-amplitude comparator responsive to said silence-threshold comparator,

a constructor of a tree with binary-valued nodes responsive to said peak-amplitude comparator, and

a pruner of tree nodes responsive to said constructor.

58. An encoder according to claim 55 wherein at least one of said one or more analysis-window functions comprises a first partial window which starts at or near a value of zero and rises to an ending value at or near one and a second partial window which starts at or near a value of one and falls to an ending value at or near zero.

59. An encoder according to claim 55 wherein said one or more discrete transforms correspond to either Evenly-Stacked Time Domain Aliasing Cancellation or Oddly-Stacked Time Domain Aliasing Cancellation, said one or more discrete transforms having a Time-Domain Aliasing Cancellation phase term responsive to said frame control.

60. A decoder for the decoding of a coded signal representing a discrete time signal and particularly a music signal, comprising

a frame control responsive to said coded signal comprising a discrete transform control, and

an inverse filter bank responsive to said coded signal and to said frame control, said inverse filter bank comprising one or more inverse discrete transforms.

61. A decoder for the decoding of a coded signal representing a discrete time signal and particularly a music signal, comprising

a frame control responsive to said coded signal comprising at least one of a discrete transform control and a synthesis window control,

an inverse filter bank responsive to said coded signal and optionally responsive to said frame control, said inverse filter bank comprising one or more inverse discrete transforms, and

a synthesis window responsive to said inverse filter bank and optionally responsive to said frame control, said synthesis window comprising one or more synthesis-window functions,

wherein at least one of said inverse filter bank and said synthesis window is responsive to said frame control.

62. A decoder according to claim 61 wherein at least one of said one or more synthesis-window functions comprises a first partial window which starts at or near a value of zero and rises to an ending value at or near one and a second partial window which starts at or near a value of one and falls to an ending value at or near zero.

63. A decoder according to claim 61 wherein said one or more inverse discrete transforms substantially correspond to either Evenly-Stacked Time Domain Aliasing Cancellation or Oddly-Stacked Time Domain Aliasing Cancellation, said one or more inverse discrete transforms having a Time-Domain Aliasing Cancellation phase term responsive to said frame control.

64. An encoder for the encoding of samples representing a discrete time signal and particularly a music signal, comprising

control means for grouping said samples into sample blocks having a length equal to a sample block length,

analysis-window means for generating analysis-window weighted blocks by weighting said sample blocks with one or more analysis-window functions, and

transform means for generating spectral information by applying a discrete transform to said analysis-window weighted blocks,

wherein said control means comprises a transform adapting means responsive to one or more characteristics of said discrete time signal for selecting said discrete transform from a plurality of discrete transforms and/or adapting said discrete transform which is a signal-independent transform.

65. A decoder for the decoding of a coded signal comprising representations of spectral components of a discrete time signal and particularly a music signal, wherein said decoder comprises

inverse control means for grouping said representations of spectral components into spectral blocks having a length equal to a spectral block length, and

transform means for generating sample blocks by applying an inverse discrete transform to said spectral blocks,

wherein said inverse control means comprises a transform adapting means responsive to said coded signal for selecting said inverse discrete transform from a plurality of inverse discrete transforms and/or adapting said inverse discrete transform which is a signal-independent transform.