Low bit-rate high-resolution spectral envelope coding for audio encoder and decoder

• US 5,581,653 A
• Filed: 08/31/1993
• Issued: 12/03/1996
• Est. Priority Date: 08/31/1993
• Status: Expired due to Term

• Current Assignee: Dolby Laboratories Licensing Corporation (Dolby Laboratories Incorporated)
• Original Assignee: Dolby Laboratories Licensing Corporation (Dolby Laboratories Incorporated)
• Inventors: Todd, Craig C.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): CHOWDHURY, INDRINAL
• Application Number:
  US08/115,513
• Time in Patent Office: 1,190 Days
• Field of Search: 395/2.14, 395/2.12, 395/2.15, 395/2.10, 395/2.74, 395/2.77, 395/2.38, 395/2.39, 395/2.33, 395/2.34
• US Class Current: 704/229
• CPC Class Codes: G10L 19/035 Scalar quantisation
• CPC Class Codes: H04B 1/667 using a division in frequency subbands for TV signals H04N19/63

Abstract

A split-band encoder prepares an estimate of an input signal spectral envelope by splitting the input signal into frequency subband signals, generates a scaled representation of the subband signals comprising scaling factors and scaled values, generates a differential coded representation of the scaling factors, and assembles the differential coded representation and scaled values into an encoded signal. The scale factors represent a spectral envelope of the input signal and spectral leakage between filterbank subbands limits the change in value between adjacent scaling factors. This limitation in change can be exploited to reduce the informational requirements of the differential coded representation. Adaptive selection between high- and low-resolution spectral envelopes and adaptive scaling factor reuse are described.

Claims

1. An encoder of one or more channels of input signals comprising signal samples, said encoder comprising

splitband means for generating a subband signal block comprising a plurality of subband signals by applying a filter bank to said channels of input signals, each subband signal representing a respective frequency subband of said channels of input signals,

scaling means for generating scaled representations of two or more subband signals, each of said scaled representations comprising a scaling factor associated with one or more scaled values, wherein said scaling means comprises means for adjusting the value of said scaling factor and said scaled values in a respective one of said scaled representations to restrict the magnitude of a difference between said scaling factor and a scaling factor in another or said scaled representations in accordance with frequency selectivity characteristics of said filter bank,

differentiating means for generating differential coded representations of one or more of said scaling factors, and

formatting means for generating an encoded signal comprising said scaled values and said differential coded representations.

2. An encoder according to claim 1 wherein said filter bank is implemented by one or more discrete transforms.

3. An encoder according to claim 1 wherein said filter bank is implemented by a bank of digital band-pass filters.

4. An encoder according to claim 1 wherein each of said subband signals represents a respective frequency subband of a respective one of said channels of input signals.

5. An encoder of two or more channels of input signals according to claim 1 wherein at least one of said subband signals represents a composite of a respective frequency subband of two or more of said channels of input signals.

6. An encoder according to claim 1 wherein each scaling factor is associated with a single scaled value.

7. An encoder according to claim 1 further comprising means for determining the information requirements of said encoded signal which results from the use of various coding modes, and then adaptively switching to the mode which achieves the greatest reduction in information requirements, said coding modes including various ranges to restrict said magnitude and various numbers of scaled values associated with a respective scale factor.

8. An encoder according to claim 1 wherein said scaling means comprises

means for generating a tentative set of scaling factors corresponding to subband signals in said subband signal block,

means for storing a previous set of scaling factors,

means for comparing results from using the scaling factors in said tentative set with results from using respective scaling factors in said previous set to generate scaled representations, and

means for selecting scaling factors from said previous set whenever the use of respective scaling factors in said tentative set results in higher information requirements for said encoded signal and/or the coding quality of said encoded signal is lower than is achieved by the use of scaling factors selected from said previous set, otherwise selecting scaling factors from said tentative set,

wherein said scaling means generates said scaled representations using said selected scaling factors, and wherein said encoded signal comprises an indication of which scaling factors are selected from said previous set.

9. An encoder according to claim 1 wherein said splitband means further comprises

means for generating one or more additional subband signal blocks each comprising a plurality of subband signals by applying a filter bank to said channels of input signals, wherein said subband signal block and said one or more additional subband signal blocks constitute a frame of blocks,

wherein said scaling means generates said scaled representations using selected scaling factors, said scaling means comprising

means for establishing sets of tentative scaling factors, one set for each block in said frame of blocks,

means for analyzing said sets of tentative scaling factors and for establishing a set of frame scaling factors in response thereto,

means for comparing, for each block in said frame of blocks, results from using said frame scaling factors with results from using respective tentative scaling factors to generate scaled representations,

means for selecting, for each block in said frame of blocks, said frame scaling factors whenever the use of said frame scaling factors results in lower information requirements of said encoded signal and/or the coding quality of said encoded signal is higher, otherwise selecting respective tentative scaling factors, and

means for generating said scaled representations using scale factors selected by said means for selecting, and

wherein said encoded signal comprises

a differential coded representation of respective tentative scaling factors in portions of said encoded signal corresponding to subband signal blocks from which scaled representations were generated using respective tentative scaling factors,

a differential coded representation of said frame scaling factors in a portion of said encoded signal corresponding to a block in said frame of blocks from which scaled representations were first generated using said frame scaling factors, and

an indication to reuse said frame scaling factors in portions of said encoded signal corresponding to one or more other blocks in said frame of blocks from which scaled representations were generated using said frame scaling factors.

10. An encoder according to claim 1 wherein said scaled representations are floating-point representations comprising exponents and mantissas, said scaling factor corresponding to an exponent and said scaled values corresponding to mantissas.

11. A decoder of an encoded signal generated by an encoder comprising an analysis filter bank, said decoder comprising

deformatting means for extracting from said encoded signal one or more scaled values and differential coded representations of one or more scale factors,

integrating means for generating scaled representations in response to said one or more scaled values and said differential coded representations, each scaled representation comprising a scaling factor associated with one or more scaled values, wherein each differential coded representation represents a difference between a respective scaling factor and a base value or another scaling factor and said integrating means is limited to integrating differences restricted in accordance with spectral selectivity of said analysis filter bank,

inverse scaling means for generating a subband signal block comprising a plurality of subband signals, wherein one or more of said subband signals is generated in response to a respective one of said scaled representations, and

inverse splitband means for generating one or more channels of output signals by applying to said subband signal block a signal synthesis filter bank inverse to said signal analysis filter bank.

12. A decoder according to claim 11 wherein said signal synthesis filter bank is implemented by one or more discrete inverse transforms.

13. A decoder according to claim 11 wherein said signal synthesis filter bank is implemented by a bank of digital inverse filters.

14. A decoder according to claim 11 wherein each of said subband signals represents a respective frequency subband of a respective one of said channels of output signals.

15. A decoder according to claim 11 wherein at least one of said subband signals represents a composite of a respective frequency subband of two or more of said channels of output signals.

16. A decoder according to claim 11 wherein said respective scaling factor is associated with a single scaled value.

17. A decoder according to claim 11 further comprising means for adaptively switching to one of a plurality of modes for generating said sealed representations, said modes including various numbers of scaled values associated with a respective scale factor.

18. A decoder according to claim 11 wherein said differential coded representations comprise an indication of scaling factor reuse, and wherein said integrating means comprises

means for storing one or more previous sets of scaling factors, and

means for incorporating one or more scaling factors from one of said previous sets into said scaled representations in response to said indication of scaling factor reuse.

19. A decoder according to claim 11 wherein said scaled representations are floating-point representations comprising exponents and mantissas, said scaling factor corresponding to an exponent and said scaled values corresponding to mantissas.

20. An encoding method of one or more channels of input signals comprising signal samples, said encoding method comprising

generating a subband signal block comprising a plurality of subband signals by applying a filter bank to said channels of input signals, each subband signal representing a respective frequency subband of said channels of input signals,

generating scaled representations of two or more subband signals, each of said scaled representations comprising a scaling factor associated with one or more scaled values, wherein said generating scaled representations comprises adjusting the value of said scaling factor and said scaled values in a respective one of said scaled representations to restrict the magnitude of a difference between said scaling factor and a scaling factor in another of said scaled representations in accordance with frequency selectivity characteristics of said filter bank such that information requirements of an encoded signal are reduced,

generating differential coded representations of one or more of said scaling factors, and

generating an encoded signal comprising said scaled values and said differential coded representations.

21. An encoding method according to claim 20 wherein said filter bank is implemented by one or more discrete transforms.

22. An encoding method according to claim 20 wherein said filter bank is implemented by a bank of digital band-pass filters.

23. An encoding method according to claim 20 wherein each of said subband signals represents a respective frequency subband of a respective one of said channels of input signals.

24. An encoding method of two or more channels of input signals according to claim 20 wherein at least one of said subband signals represents a composite of a respective frequency subband of two or more of said channels of input signals.

25. An encoding method according to claim 20 wherein each scaling factor is associated with a single scaled value.

26. An encoding method according to claim 20 further comprising determining the information requirements of said encoded signal which results from the use of various coding modes, and then adaptively switching to the mode which achieves the greatest reduction in information requirements, said coding modes including various ranges to restrict said magnitude and various numbers of sealed values associated with a respective scale factor.

27. An encoding method according to claim 20 wherein said generating scaled representations comprises

generating a tentative set of scaling factors corresponding to subband signals in said subband signal block,

storing a previous set of scaling factors,

comparing results from using the scaling factors in said tentative set with results from using respective scaling factors in said previous set to generate scaled representations, and

selecting scaling factors from said previous set whenever the use of respective scaling factors in said tentative set results in higher information requirements for said encoded signal and/or the coding quality of said encoded signal is lower than is achieved by the use of scaling factors selected from said previous set, otherwise selecting scaling factors from said tentative set,

wherein said scaled representations use said selected scaling factors, and wherein said encoded signal comprises an indication of which scaling factors are selected from said previous set.

28. An encoding method according to claim 20 wherein said generating a subband signal block further comprises

generating one or more additional subband signal blocks each comprising a plurality of subband signals by applying a filter bank to said channels of input signals, wherein said subband signal block and said one or more additional subband signal blocks constitute a frame of blocks,

wherein said generating scaled representations generates said scaled representations using selected scaling factors, said generating comprising

establishing sets of tentative scaling factors, one set for each block in said frame of blocks,

analyzing said sets of tentative scaling factors and establishing a set of frame scaling factors in response thereto,

comparing, for each block in said frame of blocks, results from using said frame scaling factors with results from using respective tentative scaling factors to generate scaled representations,

selecting, for each block in said frame of blocks, said frame scaling factors whenever the use of said frame scaling factors results in lower information requirements of said encoded signal and/or the coding quality of said encoded signal is higher, otherwise selecting respective tentative scaling factors, and

generating said scaled representations using scale factors selected by said selecting, and

wherein said encoded signal comprises

a differential coded representation of respective tentative scaling factors in portions of said encoded signal corresponding to subband signal blocks from which scaled representations were generated using respective tentative scaling factors,

a differential coded representation of said frame scaling factors in a portion of said encoded signal corresponding to a block in said frame of blocks from which scaled representations were first generated using said frame scaling factors, and

an indication to reuse said frame scaling factors in portions of said encoded signal corresponding to one or more other blocks in said frame of blocks from which scaled representations were generated using said frame scaling factors.

29. An encoding method according to claim 20 wherein said scaled representations are floating-point representations comprising exponents and mantissas, said scaling factor corresponding to an exponent and said scaled values corresponding to mantissas.

30. A decoding method for decoding an encoded signal generated by an encoder comprising an analysis filter bank, said decoding method comprising

extracting from said encoded signal one or more scaled values and differential coded representations of one or more scale factors, wherein said extracting is limited to extracting differential coded representations in which magnitudes of differentials are restricted in accordance with spectral selectivity of said analysis filter bank,

generating scaled representations in response to said one or more scaled values and said differential coded representations, each scaled representation comprising a scaling factor associated with one or more scaled values, wherein each differential coded representation represents a difference between a respective scaling factor and a base value or another scaling factor,

generating a subband signal block comprising a plurality of subband signals, wherein one or more of said subband signals is generated in response to a respective one of said scaled representations, and

generating one or more channels of output signals by applying to said subband signal block a signal synthesis filter band inverse to said signal analysis filter bank.

31. A decoding method according to claim 30 wherein said signal synthesis filter bank is implemented by one or more discrete inverse transforms.

32. A decoding method according to claim 30 wherein said signal synthesis filter bank is implemented by a bank of digital inverse filters.

33. A decoding method according to claim 30 wherein each of said subband signals represents a respective frequency subband of a respective one of said channels of output signals.

34. A decoding method according to claim 30 wherein at least one of said subband signals represents a composite of a respective frequency subband of two or more of said channels of output signals.

35. A decoding method according to claim 30 wherein said respective scaling factor is associated with a single scaled value.

36. A decoding method according to claim 30 further comprising adaptively switching to one of a plurality of modes for generating said scaled representations, said modes including various numbers of scaled values associated with a respective scale factor.

37. A decoding method according to claim 30 wherein said differential coded representations comprise an indication of scaling factor reuse, and wherein said generating a subband signal block comprises

storing one or more previous sets of scaling factors, and

incorporating one or more scaling factors from one of said previous sets into said scaled representations in response to said indication of scaling factor reuse.

38. A decoding method according to claim 30 wherein said scaled representations are floating-point representations comprising exponents and mantissas, said scaling factor corresponding to an exponent and said scaled values corresponding to mantissas.