Audio compression and decompression employing subband decomposition of residual signal and distortion reduction

• US 6,263,312 B1
• Filed: 03/02/1998
• Issued: 07/17/2001
• Est. Priority Date: 10/03/1997
• Status: Expired due to Fees

• Current Assignee: XVD Technology Holdings Ltd.
• Original Assignee: Alaris, Inc., GT Technologies
• Inventors: Troyanovsky, Boris, Ovsyannikov, Eugene, Trofimov, Andrei N., Kolesnik, Victor D., Kudryashov, Boris D., Bocharova, Irina E.
• Primary Examiner(s): Dorvil, Richemond
• Application Number:
  US09/033,431
• Time in Patent Office: 1,233 Days
• Field of Search: 704/500, 704/229, 704/501, 704/502, 704/503, 704/504, 704/200, 704/201, 704/205, 704/206, 704/212, 704/222, 704/268, 704/269, 704/227, 704/230
• US Class Current: 704/500
• CPC Class Codes: G10L 19/0208 Subband vocoders
• CPC Class Codes: G10L 19/0212 using orthogonal transformation
• CPC Class Codes: G10L 19/04 using predictive techniques

Abstract

A method and apparatus to achieve relatively high quality audio data compression/decompression, while achieving relatively low bit rates (e.g., high compression ratios). According to one aspect of the invention, a residual signal is subband decomposed and adaptively quantized and encoded to capture frequency information that may provide higher quality compression and decompression relative to transform encoding techniques. According to a second aspect of the invention, an input audio signal is compared to an encoded signal based on the input audio signal to detect and reduce, as necessary, distortion in the encoded signal or portions thereof.

Claims

1. A computer-implemented method for compressing audio data, comprising:

encoding a first frame of an input audio signal to generate a first encoded signal;

generating a first synthesized signal from the first encoded signal;

generating a first residual signal representing a difference between the first frame of the input audio signal and the first synthesized signal;

wavelet decomposing the first residual signal into a first set of residual signal subbands; and

encoding at least certain subbands in the first set of residual signal subbands.

2. The method of claim 1, wherein said encoding at least certain subbands in the first set of residual signal subbands includes:

performing a trellis quantization of at least certain subbands in the first set of residual signal subbands.

3. The method of claim 1, wherein said encoding the first frame of the input audio signal to generate the first encoded signal includes:

transform encoding the first frame of the input audio signal to generate a first set of encoded transform coefficients.

4. The method of claim 1, wherein the wavelet decomposing the first residual signal into the first set of residual signal subbands includes:

performing one or more wavelet decompositions.

5. The method of claim 1, further comprising:

encoding a second frame of the input audio signal to generate a second encoded signal;

generating a second synthesized signal from the second encoded signal;

decomposing the second synthesized signal into a second set of subbands;

decomposing the second frame of the input audio signal into a third set of subbands;

comparing at least certain parts of at least certain corresponding subbands in the second and third sets of subbands;

suppressing at least parts of the second set of subbands based on said comparing to generate a modified second set of subbands;

generating a second set of residual signal subbands representing a difference between the third set of subbands and the modified second set of subbands;

encoding at least certain subbands in the second set of residual signal subbands.

6. The method of claim 5, further comprising:

determining that the first synthesized signal is sufficiently similar to the first frame of the input audio signal prior to said step of encoding at least certain subbands in the first set of residual signal subbands; and

determining that the second synthesized signal is sufficiently dissimilar to the second frame of the input audio signal prior to said encoding at least certain subbands in the second set of residual signal subbands; and

determining to encode the first and second frames of the input audio signal differently based on said determining that the first synthesized signal is sufficiently similar and said determining that the second synthesized signal is sufficiently dissimilar.

7. The method of claim 6, wherein said determining that the second synthesized signal is sufficiently dissimilar includes:

comparing corresponding subframes of the second synthesized signal and the second frame of the input audio signal to detect distortion; and

detecting that the distortion is sufficiently high in a sufficiently large number of the subframes.

8. The method of claim 7, wherein said comparing includes:

determining a ratio between signal and noise in the subframes.

9. The method of claim 5, wherein:

said comparing includes comparing corresponding subband subframes of the second and third sets of subbands to detect distortion; and

said suppressing at least parts of the second set of subbands based on said comparing to generate the modified second set of subbands includes suppressing those subband subframes in the second set of subbands for which there is a sufficient amount of distortion detected.

10. A machine readable medium having stored thereon sequences of instructions, which when executed by a processor, cause the processor to perform the following:

encoding a first frame of an input audio signal to generate a first encoded signal;

generating a first synthesized signal from the first encoded signal;

generating a first residual signal representing a difference between the first frame of the input audio signal and the first synthesized signal;

wavelet decomposing the first residual signal into a first set of residual signal subbands; and

encoding at least certain subbands in the first set of residual signal subbands.

11. The machine readable medium of claim 10, wherein said encoding at least certain subbands in the first set of residual signal subbands includes:

performing a trellis quantization of at least certain of the first set of residual signal subbands.

12. The machine readable medium of claim 10, wherein said encoding the first frame of the input audio signal to generate the first encoded signal includes:

transform encoding the first frame of the input audio signal to generate a first set of encoded transform coefficients.

13. The machine readable medium of claim 10, wherein the wavelet decomposing the first residual signal into the first set of residual signal subbands includes:

performing one or more wavelet decompositions.

14. The machine readable medium of claim 10, further comprising:

encoding a second frame of the input audio signal to generate a second encoded signal;

generating a second synthesized signal from the second encoded signal;

decomposing the second synthesized signal into a second set of subbands;

decomposing the second frame of the input audio signal into a third set of subbands;

comparing at least certain parts of at least certain corresponding subbands in the second and third sets of subbands;

suppressing at least parts of the second set of subbands based on said step of comparing to generate a modified second set of subbands;

generating a second set of residual signal subbands representing a difference between the third set of subbands and the modified second set of subbands;

encoding at least certain subbands in the second set of residual signal subbands.

15. The machine readable medium of claim 14, further comprising:

determining that the first synthesized signal is sufficiently similar to the first frame of the input audio signal prior to said step of encoding at least certain subbands in the first set of residual signal subbands; and

determining that the second synthesized signal is sufficiently dissimilar to the second frame of the input audio signal prior to said encoding at least certain subbands in the second set of residual signal subbands; and

determining to encode the first and second frames of the input audio signal differently based on said determining that the first synthesized signal is sufficiently similar and said determining that the second synthesized signal is sufficiently dissimilar.

16. The machine readable medium of claim 15, wherein said determining that the second synthesized signal is sufficiently dissimilar includes:

comparing corresponding subframes of the second synthesized signal and the second frame of the input audio signal to detect distortion; and

detecting that the distortion is sufficiently high in a sufficiently large number of the subframes.

17. The machine readable medium of claim 16, wherein said comparing includes:

determining a ratio between signal and noise in the subframes.

18. The machine readable medium of claim 14, wherein:

said comparing includes comparing corresponding subband subframes of the second and third sets of subbands to detect distortion; and

said suppressing at least parts of the second set of subbands based on said comparing to generate the modified second set of subbands includes suppressing those subband subframes in the second set of subbands for which there is a sufficient amount of distortion detected.

19. An apparatus to compress audio data, comprising:

an encoding unit comprising an input coupled to receive an input audio signal and an output to provide an encoded signal;

a synthesizing unit coupled to the output of the encoding unit;

a first subtraction unit having inputs coupled to the output of the encoding unit and the synthesizing unit to generate a residual signal;

a residual signal wavelet decomposition unit coupled to the output of the subtraction unit to decompose the residual signal into a set of subbands; and

an quantization unit coupled to receive at least certain of the set of subbands.

20. The apparatus of claim 19, wherein the encoding unit comprises a transform encoding unit.

21. The apparatus of claim 19, wherein the quantization unit includes a trellis quantization unit to adaptively quantize at least certain of the set of subbands.

22. The apparatus of claim 19, further comprising:

an input audio signal subband decomposition unit coupled to receive the input audio signal;

a synthesized signal subband decomposition unit coupled to the output of the synthesizing unit;

a distortion reduction unit coupled to the output of the input audio signal subband decomposition unit and the synthesized signal subband decomposition unit;

a second subtraction unit having inputs coupled to the output of the distortion reduction unit and the output of the input audio signal subband decomposition unit;

a distortion detection unit coupled to receive the input audio signal and coupled to the output of the synthesizing unit to detect distortion in different frames of the synthesized signal based on comparing corresponding frames of the synthesized signal and the input audio signal, said distortion detection unit to selectively provide the output of either the residual signal subband decomposition unit or the second subtraction unit based on the level of distortion detected.

23. A computer-implemented method of compressing an input audio signal comprising:

encoding a first frame of the input audio signal to generate a first encoded signal;

generating a first synthesized signal from the first encoded signal;

decomposing the first synthesized signal into a first set of subbands;

decomposing the first frame of the input audio signal into a second set of subbands;

comparing at least certain parts of at least certain corresponding subbands in the first and second sets of subbands;

suppressing at least parts of the first set of subbands based on said step of comparing to generate a modified first set of subbands;

generating a first set of residual signal subbands representing a difference between the second set of subbands and the modified first set of subbands;

encoding at least certain of the first set of residual signal subbands.

24. The method of claim 23, wherein said encoding at least certain of the first set of residual subbands includes;

performing a trellis quantization of the first set of residual signal subbands.

25. The method of claim 23, wherein said encoding the first frame of the input audio signal to generate the first encoded signal includes:

transform encoding the first frame of the input audio signal to generate a first set of encoded transform coefficients.

26. The method of claim 23, wherein:

said comparing includes comparing corresponding subband subframes of the first and second sets of subbands to detect distortion; and

said suppressing at least parts of the first set of subbands based on said comparing to generate the modified first set of subbands includes suppressing those subband subframes in the first set of subbands for which there is a sufficient amount of distortion detected.

27. The method of claim 23, further comprising:

determining that the first synthesized signal is not sufficiently similar to the first frame of the input audio signal prior to said encoding at least certain of the first set of residual signal subbands.

28. The method of claim 27, wherein said determining that the first synthesized signal is not sufficiently similar includes:

comparing corresponding subframes of the first synthesized signal and the first frame of the input audio signal to detect distortion; and

detecting that the distortion is sufficiently high in a sufficiently large number of the subframes.

29. The method of claim 28, wherein said comparing includes:

determining a ratio between signal and noise in the subframes.

30. The method of claim 28, further comprising:

encoding a second frame of an input audio signal to generate a second encoded signal;

generating a second synthesized signal from the second encoded signal;

determining that the second synthesized signal is sufficiently similar to the second frame of the input audio signal;

generating a second residual signal representing a difference between the second frame of the input audio signal and the second synthesized signal;

decomposing the second residual signal into a second set of residual signal subbands; and

encoding at least certain of the second set of residual signal subbands.

31. The method of claim 30, wherein said decomposing the second residual signal includes performing one or more wavelet decompositions.

32. The method of claim 23, wherein said acts of decomposing include performing one or more wavelet decompositions.

33. A machine readable medium having stored thereon sequences of instructions, which when executed by a processor, cause the processor to perform the following:

encoding a first frame of an input audio signal to generate a first encoded signal;

generating a first synthesized signal from the first encoded signal;

decomposing the first synthesized signal into a first set of subbands;

decomposing the first frame of the input audio signal into a second set of subbands;

comparing at least certain parts of at least certain corresponding subbands in the first and second sets of subbands;

suppressing at least parts of the first set of subbands based on said step of comparing to generate a modified first set of subbands;

generating a first set of residual signal subbands representing a difference between the second set of subbands and the modified first set of subbands;

encoding at least certain of the first set of residual signal subbands.

34. The machine readable medium of claim 33, wherein said encoding at least certain of the first set of residual signal subbands includes:

performing a trellis quantization of the first set of residual signal subbands.

35. The machine readable medium of claim 33, wherein said encoding the first frame of the input audio signal to generate the first encoded signal includes:

transform encoding the first frame of the input audio signal to generate a first set of encoded transform coefficients.

36. The machine readable medium of claim 33, wherein:

said comparing includes the step of comparing corresponding subband subframes of the first and second sets of subbands to detect distortion; and

said suppressing at least parts of the first set of subbands based on said comparing to generate the modified first set of subbands includes suppressing those subband subframes in the first set of subbands for which there is a sufficient amount of distortion detected.

37. The machine readable medium of claim 33, further comprising:

determining that the first synthesized signal is not sufficiently similar to the first frame of the input audio signal prior to said encoding at least certain of the first set of residual signal subbands.

38. The machine readable medium of claim 37, wherein said determining that the first synthesized signal is not sufficiently similar includes:

comparing corresponding subframes of the first synthesized signal and the first frame of the input audio signal to detect distortion; and

detecting that the distortion is sufficiently high in a sufficiently large number of the subframes.

39. The machine readable medium of claim 38, wherein said comparing includes:

determining a ratio between signal and noise in the subframes.

40. The machine readable medium of claim 38, further comprising:

encoding a second frame of an input audio signal to generate a second encoded signal;

generating a second synthesized signal from the second encoded signal;

determining that the second synthesized signal is sufficiently similar to the second frame of the input audio signal;

generating a second residual signal representing a difference between the second frame of the input audio signal and the second synthesized signal;

decomposing the second residual signal into a second set of residual signal subbands; and

encoding at least certain of the second set of residual signal subbands.

41. The machine readable medium of claim 40, wherein said decomposing the second residual signal includes performing one or more wavelet decompositions.

42. The machine readable medium of claim 33, wherein said acts of decomposing include performing one or more wavelet decompositions.

43. An apparatus to compress audio data comprising:

an encoding unit comprising an input coupled to receive an input audio signal and an output to provide an encoded signal;

a synthesizing unit coupled to the output of the encoding unit;

an input audio signal subband decomposition unit coupled to receive the input audio signal;

a synthesized signal subband decomposition unit coupled to the output of the synthesizing unit;

a distortion reduction unit coupled to the output of the input audio signal subband decomposition unit and the synthesized signal subband decomposition unit;

a first subtraction unit having inputs coupled to the output of the distortion reduction unit and the output of the input audio signal wavelet decomposition unit;

a quantization unit coupled to the output of the first subtraction unit.

44. The apparatus of claim 43, wherein the encoding unit comprises a transform encoding unit.

45. The apparatus of claim 43, wherein the encoding unit includes a trellis quantization unit to adaptively quantize the set of subbands.

46. The apparatus of claim 43, wherein both the input audio signal subband decomposition unit and the synthesized signal subband decomposition unit comprise a set of wavelet filters to decompose signals into at least a high frequency subband and a low frequency subband.

47. The apparatus of claim 46, further comprising:

a second subtraction unit having inputs coupled to the output of the encoding unit and the synthesizing unit to generate a residual signal;

a residual signal subband decomposition unit coupled to the output of the subtraction unit to decompose the residual signal into a set of subbands; and

a distortion detection unit coupled to receive the input audio signal and coupled to the output of the synthesizing unit to detect distortion in different frames of the synthesized signal based on comparing corresponding frames of the synthesized signal and the input audio signal, said distortion detection unit to select the output of either the residual signal subband decomposition unit or the first subtraction unit based on the level of distortion detected.

48. A computer-implemented method of decompressing an audio signal that was compressed, said method comprising:

decompressing a first transform encoded frame to generate a first synthesized signal frame;

decompressing residual signal data associated with the first frame to generate a first set of residual signal subbands, the residual signal data representing the difference between the first frame of the original audio signal and the first transform encoded frame;

wavelet reconstructing the first set of residual signal subbands using wavelets to generate a first synthesized residual signal frame; and

adding the first synthesized signal frame and the first synthesized residual signal frame to generate a first decoded audio signal frame.

49. The method of claim 48, wherein the decompressing a first transform encoded frame to generate a first synthesized signal frame includes:

dequantizing and inverse transform coding said first transform encoded frame;

subband decomposing the result of said step of dequantizing and inverse transform coding to generate a first set of subbands;

inspecting the input data to determine which parts of the subbands were suppressed during compression of the original audio signal;

suppressing those parts of the first set of subbands; and

subband reconstructing the results of said step of suppressing.

50. The method of claim 49, wherein said subband decomposing and said subband reconstructing include respectively performing one or more wavelet decompositions and reconstructions.

51. The method of claim 48 wherein:

said decompressing the first transform encoded frame to generate the first synthesized signal frame includes,

dequantizing and inverse transform coding said first transform encoded frame to generate said first synthesized signal frame; and

said method further includes,

decoding a second transform encoded frame to generate a second synthesized signal frame;

subband decomposing the second synthesized signal frame into a first set of synthesized signal subbands;

suppressing those parts of the first set of synthesized signal subbands that were suppressed during compression;

decoding residual signal data associated with the second frame to generate a second set of residual signal subbands, the residual signal data representing the difference between the second frame of the original audio signal and the second transform encoded frame;

subband reconstructing the second set of residual signal subbands to generate a second synthesized residual signal frame; and

adding the second synthesized signal frame and the second synthesized residual signal frame to generate a second decoded audio signal frame.

52. A machine readable medium having stored thereon sequences of instructions, which when executed by a processor, cause the processor to perform the following:

decompressing a first transform encoded frame to generate a first synthesized signal frame;

decompressing residual signal data associated with the first frame to generate a first set of residual signal subbands, the residual signal data representing the difference between the first frame of the original audio signal and the first transform encoded frame;

wavelet reconstructing the first set of residual signal subbands using wavelets to generate a first synthesized residual signal frame; and

adding the first synthesized signal frame and the first synthesized residual signal frame to generate a first decoded audio signal frame.

53. The machine readable medium of claim 52, wherein the decompressing a first transform encoded frame to generate a first synthesized signal frame includes:

dequantizing and inverse transform coding said first transform encoded frame;

subband decomposing the result of said dequantizing and inverse transform coding to generate a first set of subbands;

inspecting the input data to determine which parts of the subbands were suppressed during compression of the original audio signal;

suppressing those parts of the first set of subbands; and

subband reconstructing the results of said suppressing.

54. The machine readable medium of claim 53, wherein said subband decomposing and said subband reconstructing include respectively performing one or more wavelet decompositions and reconstructions.

55. The machine readable medium of claim 52 wherein:

said decompressing the first transform encoded frame to generate the first synthesized signal frame includes,

dequantizing and inverse transform coding said first transform encoded frame to generate said first synthesized signal frame; and

said method further includes,

decoding a second transform encoded frame to generate a second synthesized signal frame;

subband decomposing the second synthesized signal frame into a first set of synthesized signal subbands;

suppressing those parts of the first set of synthesized signal subbands that were suppressed during compression;

decoding residual signal data associated with the second frame to generate a second set of residual signal subbands, the residual signal data representing the difference between the second frame of the original audio signal and the second transform encoded frame;

subband reconstructing the second set of residual signal subbands to generate a second synthesized residual signal frame; and

adding the second synthesized signal frame and the second synthesized residual signal frame to generate a second decoded audio signal frame.

56. A computer-implemented method of decompressing an audio signal that was compressed, said method comprising:

decompressing a first transform encoded frame into a first synthesized signal frame;

subband decomposing the first synthesized signal frame into a first set of synthesized signal subbands;

suppressing those parts of the first set of synthesized signal subbands that were suppressed during compression;

subband reconstructing the results of the suppressing to generate a first distortion-reduced synthesized signal frame;

decompressing residual signal data associated with the first frame to generate a first set of residual signal subbands, the residual signal data representing the difference between the first frame of the original audio signal and the first transform encoded frame;

subband reconstructing the first set of residual signal subbands to generate a first synthesized residual signal frame; and

adding the first distortion-reduced synthesized signal frame and the first synthesized residual signal frame to generate a first decompressed audio signal frame.

57. The method of claim 56, wherein said subband decomposing and the subband reconstructing are performed using wavelets.

58. The method of claim 56, wherein said decompressing residual signal data includes:

performing a trellis dequantization.

59. The method of claim 56, further comprising:

decompressing a second transform encoded frame to generate a second synthesized signal frame;

decompressing residual signal data associated with the second frame to generate a second set of residual signal subbands, the residual signal data representing the difference between the second frame of the original audio signal and the second transform encoded frame;

subband reconstructing the second set of residual signal subbands using wavelets to generate a second synthesized residual signal frame; and

adding the second synthesized signal frame and the second synthesized residual signal frame to generate a second decompressed audio signal frame.

60. A machine readable medium having stored thereon sequences of instructions, which when executed by a processor, cause the processor to perform the following:

decompressing a first transform encoded frame into a first synthesized signal frame;

subband decomposing the first synthesized signal frame into a first set of synthesized signal subbands;

suppressing those parts of the first set of synthesized signal subbands that were suppressed during compression;

subband reconstructing the results of the step of suppressing to generate a first distortion-reduced synthesized signal frame;

decompressing residual signal data associated with the first frame to generate a first set of residual signal subbands, the residual signal data representing the difference between the first frame of the original audio signal and the first transform encoded frame;

subband reconstructing the first set of residual signal subbands to generate a first synthesized residual signal frame; and

adding the first distortion-reduced synthesized signal frame and the first synthesized residual signal frame to generate a first decompressed audio signal frame.

61. The machine readable medium of claim 60, wherein said subband decomposing and the subband reconstructing are performed using wavelets.

62. The machine readable medium of claim 60, wherein said decompressing residual signal data includes:

performing a trellis dequantization.

63. The machine readable medium of claim 60, further comprising:

decompressing a second transform encoded frame to generate a second synthesized signal frame;

decompressing residual signal data associated with the second frame to generate a second set of residual signal subbands, the residual signal data representing the difference between the second frame of the original audio signal and the second transform encoded frame;

subband reconstructing the second set of residual signal subbands using wavelets to generate a second synthesized residual signal frame; and

adding the second synthesized signal frame and the second synthesized residual signal frame to generate a second decompressed audio signal frame.