Pitch determination using speech classification and prior pitch estimation

• US 6,507,814 B1
• Filed: 09/18/1998
• Issued: 01/14/2003
• Est. Priority Date: 08/24/1998
• Status: Expired due to Term

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Conexant Systems Incorporated (Synaptics Incorporated)
• Inventors: Gao, Yang
• Primary Examiner(s): Tsang, Fan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/154,654
• Time in Patent Office: 1,579 Days
• Field of Search: 704/207, 704/216, 704/217, 704/218, 704/219, 704/220
• US Class Current: 704/220
• CPC Class Codes: G10L 19/002 Dynamic bit allocation for perceptual audio coders G10L19/032
• CPC Class Codes: G10L 19/005 Correction of errors induced by the transmission channel, if related to the coding algorithm
• CPC Class Codes: G10L 19/012 Comfort noise or silence coding
• CPC Class Codes: G10L 19/083 the excitation function being an excitation gain G10L25/90 takes precedence
• CPC Class Codes: G10L 19/09 Long term prediction, i.e. removing periodical redundancies, e.g. by using adaptive codebook or pitch predictor
• CPC Class Codes: G10L 19/10 the excitation function being a multipulse excitation
• CPC Class Codes: G10L 19/12 the excitation function being a code excitation, e.g. in code excited linear prediction [CELP] vocoders
• CPC Class Codes: G10L 19/125 Pitch excitation, e.g. pitch synchronous innovation CELP [PSI-CELP]
• CPC Class Codes: G10L 19/18 Vocoders using multiple modes
• CPC Class Codes: G10L 19/265 Pre-filtering, e.g. high frequency emphasis prior to encoding
• CPC Class Codes: G10L 2019/0005 Multi-stage vector quantisation
• CPC Class Codes: G10L 2019/0007 Codebook element generation
• CPC Class Codes: G10L 2019/0011 Long term prediction filters, i.e. pitch estimation
• CPC Class Codes: G10L 21/0364 for improving intelligibility

Abstract

A multi-rate speech codec supports a plurality of encoding bit rate modes by adaptively selecting encoding bit rate modes to match communication channel restrictions. In higher bit rate encoding modes, an accurate representation of speech through CELP (code excited linear prediction) and other associated modeling parameters are generated for higher quality decoding and reproduction. To achieve high quality in lower bit rate encoding modes, the speech encoder departs from the strict waveform matching criteria of regular CELP coders and strives to identify significant perceptual features of the input signal. To support lower bit rate encoding modes, a variety of techniques are applied many of which involve the classification of the input signal. For each bit rate mode selected, pluralities of fixed or innovation subcodebooks are selected for use in generating innovation vectors. The speech encoder also utilizes an adaptive weighting factor in the selection of a current pitch lag value from a plurality of pitch lag candidates. For example, if the speech encoder identifies an integer multiple timing relationship between any two pitch lag candidates, the pitch lag candidate with the smallest timing value is favored through adjustment of the weighting factor. Similarly, if a pitch lag candidate exhibits timing that corresponds to that of previous pitch lag values, the weighting factor is adjusted to favor that candidate.

Claims

1. A speech encoding system for encoding a speech signal including a previous pitch lag and a current pitch lag, the speech encoding system comprising:

an adaptive codebook for storing excitation vectors associated with corresponding pitch lag candidates; and

an encoder processing circuit for identifying the pitch lag candidates for at least one of a frame and a sub-frame of the speech signal;

the encoder processing circuit selecting a preferential one of the pitch lag candidates as the current pitch lag based on at least two of the following;

a first timing relationship, a second timing relationship, and voiced classification;

the first timing relationships concerning a temporal relationship between the previous pitch lag and at least one of the pitch lag candidates, the second timing relationship concerning a temporal relationship between at least two of the pitch lag candidates, the voiced classification pertaining to an interval of the speech signal.

2. The speech encoding system of claim 1 wherein the second timing relationship comprises an integer multiple timing relationship between at least two of the plurality of pitch lag candidates.

3. The speech encoding system of claim 2 wherein the encoder processing circuit considers the integer multiple timing relationship in the selection of the preferential one of the pitch lag candidates.

4. The speech encoding system of claim 1 wherein the encoder processing circuit favors the selection of the preferential one of the pitch lag candidates if the at least one preferential one of the pitch lag candidates and the previous pitch lag are within a temporal neighborhood of each other.

5. The speech encoding system of claim 4 wherein favoring the selection involves application of a weighting factor to a pitch correlation value associated with at least one of the pitch lag candidates.

6. The speech encoding system of claim 4 wherein the encoder processing circuit applies a pitch correlation with reference to at least one of said timing relationships to identify the pitch lag candidates.

7. The speech encoding system of claim 6 wherein the encoder processing circuit applies the weighting factor to the pitch correlation.

34. The speech encoding system of claim 1 wherein the voiced classification pertains to a prior interval as the interval of the speech signal.

8. A speech encoding system for encoding a speech signal that has a current pitch lag, the speech encoding system comprising:

an adaptive codebook;

an encoder processing circuit that identifies a plurality of pitch lag candidates; and

the encoder processing circuit applying an adaptive weighting factor to a pitch correlation to favor selection of at least one of the pitch lag candidates over at least one other of the pitch lag candidates if at least one of a first timing relationship and a second timing relationship is detected;

the first timing relationship associated with one of the pitch lag candidates and the second timing relationship being between at least two of the pitch lag candidates;

the encoder processing circuit selecting one of the pitch lag candidates as the current pitch lag by comparing the weighted pitch correlation to another pitch correlation.

9. The speech encoding system of claim 8 wherein the encoder processing circuit adjusts the adaptive weighting factor if an integer multiple timing relationship is detected as the second timing relationship between at least two of the plurality of pitch lag candidates.

10. The speech encoding system of claim 8 wherein the speech signal has a previous pitch lag, and the encoder processing circuit adjusts the adaptive weighting factor if the first timing relationship is detected between a previous pitch lag and any one of the plurality of pitch lag candidates and if a previous speech interval is generally voiced.

11. The speech encoding system of claim 9 wherein the speech signal has previous pitch lag, and the encoder processing circuit also adjusts the adaptive weighting factor if the first timing relationship is detected between a previous pitch lag and any one of the plurality of pitch lag candidates and if at least one previous speech signal is generally voiced.

12. The speech encoding system of claim 9 wherein the encoder processing circuit applies correlation to identify the plurality of pitch lag candidates.

13. The speech encoding system of claim 10 wherein the encoder processing circuit applies correlation to identify the plurality of pitch lag candidates.

14. The speech encoding system of claim 12 wherein the encoder applies the adaptive weighting factor with the correlation.

15. The speech encoding system of claim 12 wherein the encoder applies the adaptive weighting factor with the correlation.

35. The speech encoding system of claim 8 wherein the weighting factor is adjusted based on satisfaction of at least one of said timing relationships.

36. The speech encoding system of claim 8 where a presence of a generally voiced prior interval determines a value of the adaptive weighting factor for selection of the current pitch lag.

16. A method for speech encoding, the method comprising:

identifying a plurality of pitch lag candidates;

using an adaptive weighting factor applied to a pitch correlation to favor at least one of the pitch lag candidates over at least one other of the pitch lag candidates if at least one of a first timing relationship and a second timing relationship is detected;

the first timing relationship associated with one of the pitch lag candidates and the second timing relationship being between at least two of the pitch lag candidates; and

selecting one of the plurality of the pitch lag candidates as a current pitch lag estimate by comparing the weighted pitch correlation to another pitch correlation.

17. The method of claim 16 further comprising adjusting the adaptive weighting factor if an integer multiple timing relationship is detected as the second timing relationship between at least two of the plurality of pitch lag candidates.

18. The method of claim 16 wherein the speech signal has a previous pitch lag, and further comprising adjusting the adaptive weighting factor if the first timing relationship is detected between the previous pitch lag and any one of the plurality of pitch lag candidates and if a previous speech interval is generally voiced.

19. The method of claim 17 wherein the speech signal has a previous pitch lag, and further comprising also adjusting the adaptive weighting factor if the first timing relationship is detected between the previous pitch lag and any one of the plurality of pitch lag candidates and if at least a previous speech interval is generally voiced.

20. The speech encoding system of claim 16 wherein the identifying the plurality of pitch lag candidates involves application of correlation to which the adaptive weighting factor is applied.

37. The speech encoding system of claim 16 where a presence of a generally voiced prior interval determines a value of the adaptive weighting factor for selection of the current pitch lag.

21. A method of encoding a speech signal, the method comprising the steps of:

identifying a plurality of pitch lag candidates for a present interval of the speech signal;

determining if a previous interval, with respect to the present interval, contains a voiced component;

comparing the identified pitch lag candidates to at least one previous pitch lag value for a previous interval;

to identify at least one favored one of the pitch lag candidates that falls within a temporal neighborhood of the previous pitch lag value if the previous interval contains a generally voiced component; and

favoring selection of the at least one favored one of the pitch lag candidates as a preferential one of the pitch lag candidates by weighting a pitch correlation for at least one favored candidate differently than a remainder of the pitch lag candidates.

22. The method according to claim 21 further comprising selecting a preferential one of candidates by correlating a target signal with a synthesized signal derived with reference to the at least one favored candidate.

23. The method according to claim 21 further comprising selecting a preferential one of the candidates by correlating a target signal with a synthesized signal derived with reference to the pitch lag candidates.

24. The method according to claim 21 further comprising detecting a first timing relationship between at least one favored one of pitch lag candidates and a previous pitch lag, where the first timing relationship is present if at least one favored one of the pitch lag candidates falls within the temporal neighborhood of the previous pitch lag.

25. The method according to claim 24 further comprising the steps of:

comparing the identified pitch lag candidates to each other;

detecting a second timing relationship if the compared pitch lag candidates have pitch lags related approximately by an integer multiple of each other.

26. The method according to claim 25 further comprising the steps of:

favoring selection of the a second favored one of the pitch lag candidates with a second timing relationship as the preferential one of the pitch lag candidates by weighting the pitch correlation for the second favored one differently than a remainder of the pitch lag candidates.

27. A method of encoding a speech signal, the method comprising the steps of:

identifying a plurality of pitch lag candidates for a present interval of the speech signal;

determining if a previous interval, with respect to the present interval, contains a voiced component;

comparing identified pitch lag candidates to each other;

detecting a timing relationship if the compared pitch lag candidates have pitch lags related approximately by an integer multiple of each other; and

favoring selection of at least one favored one of the pitch lag candidates with the timing relationship as a preferential one of the pitch lag candidates by weighting a pitch correlation for the at least one favored candidate differently than a remainder of the pitch lag candidates.

28. A method of encoding a speech signal, the method comprising:

identifying a plurality of regions of the pitch lag;

determining a local maximum correlation between a target speech signal and a synthesized speech signal within each of the identified regions to provide a set of local maximum correlations; and

selecting a global maximum correlation among the determined local maximum correlations to facilitate selection of a pitch lag for a present interval of a speech signal.

29. The method according to claim 28 further comprising determining a pitch lag associated with the selected global maximum correlation as a present pitch lag if the selected global maximum correlation represents the local maximum correlation of a first or predecessor region of the regions.

30. The method according to claim 28 further comprising:

comparing the selected global maximum correlation to local maximum correlations if the selected global maximum is outside of the first or predecessor region of the regions.

31. The method according to claim 30 further comprising:

applying weighting to pitch correlation values for candidate pitch lags based on a first timing relationship reflecting a neighborhood of a preferential candidate in relation to other candidate pitch lags associated with the regions prior to the comparing step.

32. The method according to claim 31 further comprising:

applying weighting to pitch correlation values for candidate pitch lags based on a second timing relationship, modifying the values of the determined local maximum correlations prior to the comparing step.

33. The method according to claim 31 further comprising:

applying weighting to the pitch correlation values for candidate pitch lags based on both a first timing relationship reflecting a selected candidate in relation to previous pitch lag values and a second relationship reflecting a selected candidate in relation to other candidate pitch lag values.