Efficient coding of digital media spectral data using wide-sense perceptual similarity

• US 8,645,127 B2
• Filed: 11/26/2008
• Issued: 02/04/2014
• Est. Priority Date: 01/23/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Mehrotra, Sanjeev, Chen, Wei-Ge
• Primary Examiner(s): Han, Qi
• Application Number:
  US12/324,689
• Publication Number:
  US 20090083046A1
• Time in Patent Office: 1,896 Days
• Field of Search: 704/200.1, 704/500, 704/206, 704/200, 704/201, 704/205, 704/501
• US Class Current: 704/206
• CPC Class Codes: G10L 19/0204 using subband decomposition
• CPC Class Codes: G10L 19/0208 Subband vocoders
• CPC Class Codes: G10L 19/035 Scalar quantisation

Abstract

Traditional audio encoders may conserve coding bit-rate by encoding fewer than all spectral coefficients, which can produce a blurry low-pass sound in the reconstruction. An audio encoder using wide-sense perceptual similarity improves the quality by encoding a perceptually similar version of the omitted spectral coefficients, represented as a scaled version of already coded spectrum. The omitted spectral coefficients are divided into a number of sub-bands. The sub-bands are encoded as two parameters: a scale factor, which may represent the energy in the band; and a shape parameter, which may represent a shape of the band. The shape parameter may be in the form of a motion vector pointing to a portion of the already coded spectrum, an index to a spectral shape in a fixed code-book, or a random noise vector. The encoding thus efficiently represents a scaled version of a similarly shaped portion of spectrum to be copied at decoding.

Claims

1. An audio encoding method, comprising:

with a computer,

transforming an input audio signal block into a set of spectral coefficients,

dividing the spectral coefficients into plural bands,

coding values of the spectral coefficients of at least one of the bands in an output bitstream,

searching the at least one of the bands coded as spectral coefficient values for a portion similar to at least one other band of the plural bands, and

coding the at least one other band in the output bitstream as a scaled version of a shape of the portion of the at least one of the bands coded as spectral coefficient values, wherein the coding the at least one other band comprises coding the at least one other band using a scale parameter and a shape parameter, the shape parameter comprising a motion vector based on results of the searching that indicates the portion of the at least one of the bands coded as spectral coefficient values, and wherein the scale parameter is a scaling factor to scale the portion.

2. The audio encoding method of claim 1, wherein the scaling factor represents a total energy for the at least one other band.

3. The audio encoding method of claim 1, wherein the scaling factor is coded as coefficients characterizing a polynomial relation that yields scaling factors of two or more of the at least one other band as a function of frequency.

4. The audio encoding method of claim 1, wherein the scaling factor is a root-mean-square value of coefficients within the at least one other band.

5. The audio encoding method of claim 1, wherein the shape parameter further comprises values representing shift of the portion.

6. The audio encoding method of claim 1, wherein the shape parameter further comprises values representing stretch of the portion.

7. The audio encoding method of claim 1, wherein the motion vector indicates a normalized version of the portion.

8. The audio encoding method of claim 1, wherein the coding the at least one other band comprises coding the at least one other band as a filter having a frequency response and excitation.

9. The audio encoding method of claim 8, wherein the filter having the frequency response is a linear predictive coding filter.

10. The audio encoding method of claim 1, wherein the shape parameter further comprises a value that indicates a spectral shape from a codebook.

11. The audio encoding method of claim 1, further comprising:

selecting the portion of the at least one of the bands coded as spectral coefficient values by performing a least-means-square comparison of a normalized version of the at least one other band; and

storing an indication of the selected portion in the motion vector.

12. One or more computer-readable storage devices or memory comprising instructions configurable to cause a computer to perform an audio decoding method for an encoded audio bitstream, the method comprising:

decoding baseband spectral coefficients from the encoded audio bitstream;

decoding a shape parameter from the encoded audio bitstream, the shape parameter comprising a motion vector identifying one or more baseband spectral coefficients, the motion vector including a value that was set as a result of searching the baseband spectral coefficients for a portion of the baseband spectral coefficients similar to one or more extended band spectral coefficients; and

decoding the one or more extended band spectral coefficients by;

copying the one or more identified baseband spectral coefficients according to the shape parameter, and

scaling the copied one or more identified baseband spectral coefficients according to a scale parameter.

13. The one or more computer-readable storage devices or memory of claim 12, wherein the shape parameter further comprises a value that indicates a spectral shape in a codebook, and wherein the decoding one or more extended band spectral coefficients further comprises copying the spectral shape from the codebook.

14. The one or more computer-readable storage devices or memory of claim 12, wherein the scale parameter comprises a scaling factor representing a total energy of a band of spectral coefficients from which the encoded audio bitstream was encoded.

15. The one or more computer-readable storage devices or memory of claim 12, wherein the scale parameter comprises a scaling factor, the scaling factor being a root-mean-square value of a band of spectral coefficients from which the encoded audio bitstream was encoded.

16. The one or more computer-readable storage devices or memory of claim 12, wherein the audio decoding method further comprises performing an inverse transform operation to transform the decoded one or more baseband spectral coefficients and the decoded one or more extended band spectral coefficients into a reproduction of an input audio signal block.

17. The one or more computer-readable storage devices or memory of claim 12, wherein the scale parameter comprises coefficients characterizing a polynomial relation that yields scaling factors for a plurality of extended band spectral coefficients as a function of frequency.

18. A computing device comprising:

a processing unit;

one or more computer-readable storage media comprising instructions configured to cause the processing unit to perform an audio decoding method for an encoded audio bitstream, the method comprising;

decoding baseband spectral coefficients from the encoded audio bitstream;

decoding a first band of extended band spectral coefficients from the encoded audio bitstream by;

decoding, from the encoded audio bitstream, a scale factor for the first band;

copying one or more identified baseband spectral coefficients according to a first shape parameter, wherein the first shape parameter comprises a motion vector identifying one or more baseband spectral coefficients to be copied, the identified one or more baseband spectral coefficients describing a shape of a spectral band, the motion vector including a value that was set as a result of searching the baseband spectral coefficients for a portion of the baseband spectral coefficients similar to one or more of the first band of extended band spectral coefficients; and

scaling the copied one or more identified baseband spectral coefficients according to the decoded scale factor for the first band;

decoding a second band of the extended band spectral coefficients from the encoded audio bitstream by;

decoding, from the encoded audio bitstream, a scale factor for the second band;

copying one or more vectors from a codebook according to a second shape parameter; and

scaling the copied one or more vectors from the codebook according to the decoded scale factor for the second band; and

performing an inverse transform on the decoded baseband spectral coefficients and the decoded extended band spectral coefficients to make a reconstructed audio signal.

19. The computing device of claim 18, wherein the decoded scale factor for the first band comprises a root-mean-square value of a band of spectral coefficients from which the encoded audio bitstream was encoded.

20. The computing device of claim 18, wherein the first shape parameter further comprises values representing a stretch of the shape of the spectral band.

21. One or more computer-readable storage devices or memory comprising instructions configurable to cause a computer to perform an audio encoding method, the method comprising:

transforming an input audio signal block into a set of spectral coefficients,

dividing the spectral coefficients into plural bands,

coding values of the spectral coefficients of at least one of the bands in an output bitstream,

searching the at least one of the bands coded as spectral coefficient values for a portion similar to at least one other band of the plural bands, and

coding the at least one other band in the output bitstream as a scaled version of a shape of the portion of the at least one of the bands coded as spectral coefficient values, wherein the coding the at least one other band comprises coding the at least one other band using a scale parameter and a shape parameter, the shape parameter comprising a motion vector based on results of the searching that indicates the portion of the at least one of the bands coded as spectral coefficient values, and wherein the scale parameter is a scaling factor to scale the portion.

22. The computer-readable storage devices or memory of claim 21, wherein the scaling factor represents a total energy for the at least one other band.

23. The computer-readable storage devices or memory of claim 21, wherein the scaling factor is coded as coefficients characterizing a polynomial relation that yields scaling factors of two or more of the at least one other band as a function of frequency.

24. The computer-readable storage devices or memory of claim 21, wherein the scaling factor is a root-mean-square value of coefficients within the at least one other band.

25. The computer-readable storage devices or memory of claim 21, wherein the shape parameter further comprises values representing shift of the portion.

26. The computer-readable storage devices or memory of claim 21, wherein the shape parameter further comprises values representing stretch of the portion.

27. The computer-readable storage devices or memory of claim 21, wherein the motion vector indicates a normalized version of the portion.

28. The computer-readable storage devices or memory of claim 21, wherein the coding the at least one other band comprises coding the at least one other band as a filter having a frequency response and excitation.

29. The computer-readable storage devices or memory of claim 28, wherein the filter having the frequency response is a linear predictive coding filter.

30. The computer-readable storage devices or memory of claim 21, wherein the shape parameter further comprises a value that indicates a spectral shape from a codebook.

31. The computer-readable storage devices or memory of claim 21, wherein the audio encoding method further comprises:

selecting the portion of the at least one of the bands coded as spectral coefficient values by performing a least-means-square comparison of a normalized version of the at least one other band; and

storing an indication of the selected portion in the motion vector.

32. A computing device comprising:

a processing unit;

one or more computer-readable storage media comprising instructions configured to cause the processing unit to perform an audio encoding method, the method comprising;

transforming an input audio signal block into a set of spectral coefficients,

dividing the spectral coefficients into plural bands,

coding values of the spectral coefficients of at least one of the bands in an output bitstream,

searching the at least one of the bands coded as spectral coefficient values for a portion similar to at least one other band of the plural bands, and

coding the at least one other band in the output bitstream as a scaled version of a shape of the portion of the at least one of the bands coded as spectral coefficient values, wherein the coding the at least one other band comprises coding the at least one other band using a scale parameter and a shape parameter, the shape parameter comprising a motion vector based on results of the searching that indicates the portion of the at least one of the bands coded as spectral coefficient values, and wherein the scale parameter is a scaling factor to scale the portion.

33. The computing device of claim 32, wherein the scaling factor is coded as coefficients characterizing a polynomial relation that yields scaling factors of two or more of the at least one other band as a function of frequency.

34. The computing device of claim 32, wherein the scaling factor is a root-mean-square value of coefficients within the at least one other band.

35. The computing device of claim 32, wherein the coding the at least one other band comprises coding the at least one other band as a filter having a frequency response and excitation.

36. The computing device of claim 35, wherein the filter having the frequency response is a linear predictive coding filter.

37. The computing device of claim 32, wherein the shape parameter further comprises a value that indicates a spectral shape from a codebook.

38. The computing device of claim 32, wherein the audio encoding method further comprises:

selecting the portion of the at least one of the bands coded as spectral coefficient values by performing a least-means-square comparison of a normalized version of the at least one other band; and

storing an indication of the selected portion in the motion vector.