Adaptive postfiltering methods and systems for decoding speech

• US 20030088406A1
• Filed: 06/28/2002
• Published: 05/08/2003
• Est. Priority Date: 10/03/2001
• Status: Active Grant

• Current Assignee: Avago Technologies International Sales Pte Limited (Broadcom, Inc.)
• Original Assignee: Broadcom Corporation (Broadcom, Inc.)
• Inventors: Chen, Juin-Hwey, Thyssen, Jes
• Granted Patent: US 7,512,535 B2
• Time in Patent Office: Days
• US Class Current: 704/219
• CPC Class Codes: G10L 19/26 Pre-filtering or post-filtering

Abstract

A filter controller processes a decoded speech (DS) signal. The DS signal has a spectral envelope including a first plurality of formant peaks having different respective amplitudes. The controller produces, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal. The spectrally-flattened time-domain DS signal has a spectral envelope including a second plurality of formant peaks. Each of the second plurality of formant peaks approximately coincides in frequency with a respective one of the first plurality of formant peaks. Also, the second plurality of formant peaks have approximately equal respective amplitudes. Next, the controller derives, from the spectrally-flattened time-domain DS signal, a set of filter coefficients representative of a filter response that is to be used to filter the DS signal.

Claims

1. A method of processing a decoded speech (DS) signal, the DS signal having a spectral envelope including a first plurality of formant peaks, comprising:

(a) producing, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened time-domain DS signal having a spectral envelope including a second plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the second plurality of formant peaks are less than one or more corresponding amplitude differences between respective ones of the first plurality of formant peaks; and

(b) deriving a set of filter coefficients from the spectrally-flattened time-domain DS signal.

2. The method of claim 1, wherein each of the second plurality of formant peaks approximately coincides in frequency with a respective one of the first plurality of formant peaks.

3. The method of claim 1, wherein step (a) comprises filtering the DS signal to produce the spectrally-flattened time-domain DS signal.

4. The method of claim 1, wherein step (a) comprises:

(a)(i) filtering the DS signal to produce an intermediate spectrally-flattened DS signal having a spectral envelope including an intermediate plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the intermediate plurality of formant peaks are less than the one or more amplitude differences between the respective ones of the first plurality of formant peaks; and

(a)(ii) filtering the intermediate spectrally-flattened DS signal to reduce the one or more amplitude differences between the respective ones of the intermediate plurality of formant peaks, thereby producing the spectrally-flattened DS signal.

5. The method of claim 4, wherein step (a) further comprises, prior to step (a)(i):

deriving an intermediate set of filter coefficients based on the DS signal,

wherein step (a)(i) comprises filtering the DS signal using the intermediate set of filter coefficients.

6. The method of claim 5, wherein said step of deriving the intermediate set of filter coefficients includes performing an LPC analysis on the DS signal.

7. The method of claim 5, wherein said step of deriving the intermediate set of filter coefficients includes deriving the intermediate set of filter coefficients from a set of LPC predictor coefficients associated with the DS signal.

8. The method of claim 5, wherein step (a) further comprises, prior to step (a)(ii):

deriving a second intermediate set of filter coefficients from the intermediate spectrally-flattened DS signal,

wherein step (a)(ii) comprises filtering the intermediate spectrally-flattened DS signal using the second intermediate set of filter coefficients.

9. The method of claim 1, wherein step (b) comprises deriving the set of filter coefficients without performing a time-domain to a frequency-domain conversion.

10. The method of claim 1, wherein step (b) comprise performing a time-domain analysis on the spectrally-flattened time-domain DS signal.

11. The method of claim 1, wherein step (b) comprises:

performing an LPC analysis of the spectrally-flattened time-domain DS signal.

12. The method of claim 1, further comprising:

(c) bandwidth expanding the set of filter coefficients to produce a set of bandwidth expanded filter coefficients; and

(d) smoothing the bandwidth expanded set of filter coefficients to produce a smoothed set of filter coefficients.

13. The method of claim 12, further comprising:

(e) filtering the DS signal using the smoothed set of filter coefficients.

14. The method of claim 1, wherein the one or more amplitude differences between the second plurality of formant peaks are less than 3 dB when the one or more amplitude differences between the first plurality of formant peaks are less than approximately 30 dB.

15. The method of claim 1, wherein:

the set of filter coefficients represent a filter response having a plurality of spectral peaks, each spectral peak approximately coinciding in frequency with a respective one of the first plurality of formant peaks; and

one or more amplitude differences between respective ones of the spectral peaks are less than corresponding amplitude differences between respective ones of the first plurality of formant peaks.

16. The method of claim 1, wherein the set of filter coefficients are all-pole filter coefficients.

17. The method of claim 1, wherein the set of filter coefficients are one of a set of pole-zero filter coefficients and a set of all-zero filter coefficients.

18. The method of claim 1, wherein the frequency response has a spectral tilt that is reduced relative to a spectral tilt of the DS signal.

19. A method of processing a decoded speech (DS) signal, the DS signal having a spectral envelope including a first plurality of formant peaks having different respective amplitudes, comprising:

(a) producing, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened time-domain DS signal having a spectral envelope including a second plurality of formant peaks, each of the second plurality of formant peaks approximately coinciding in frequency with a respective one of the first plurality of formant peaks, the second plurality of formant peaks having approximately equal respective amplitudes; and

(b) deriving a set of filter coefficients from the spectrally-flattened time-domain DS signal.

20. The method of claim 19, wherein the respective amplitudes of the second plurality of formant peaks are within 3 dB of each other when the respective amplitudes of the first plurality of formant peaks are within an approximate range of less than 30 dB of each other.

21. A method of processing a decoded speech (DS) signal having a spectral tilt, comprising:

(a) producing, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened DS signal having a reduced spectral tilt relative to the DS signal; and

(b) deriving, from the spectrally-flattened time-domain decoded speech signal, a set of filter coefficients, representative of a filter response having a reduced spectral tilt relative to the spectral tilt of the DS signal.

22. The method of claim 21, wherein the filter response includes a plurality of spectral peaks approximately coinciding in frequency with formant peaks of the DS signal.

23. The method of claim 21, wherein step (a) comprises:

(a)(i) filtering the DS signal to reduce the spectral tilt therein, thereby producing an intermediate DS signal having a reduced spectral tilt relative to the DS signal; and

(a)(i) filtering the intermediate DS signal to reduce the spectral tilt therein, thereby producing the spectrally-flattened DS signal t(n).

24. A method of processing a decoded speech (DS) signal, the DS signal having a spectral envelope including a first plurality of formant peaks, comprising:

(a) producing, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened time-domain DS signal having a spectral envelope including a second plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the second plurality of formant peaks are less than one or more corresponding amplitude differences between respective ones of the first plurality of formant peaks;

(b) deriving, from the spectrally-flattened time-domain DS signal, a set of filter coefficients representative of a filter response, the filter response having spectral peaks corresponding to the second plurality of formant peaks; and

(c) filtering the DS signal using the set of filter coefficients.

25. A method of processing a decoded speech (DS) signal, the DS signal having a spectral envelope including a first plurality of formant peaks having different respective amplitudes, comprising:

(a) deriving a first set of filter coefficients based on the DS signal;

(b) filtering the DS signal based on the first set of filter coefficients, to produce a first filtered DS signal;

(c) deriving a second set of filter coefficients based on the first filtered DS signal;

(d) filtering the first filtered DS signal based on the second set of filter coefficients, to produce a second filtered DS signal, the second filtered DS signal having a spectral envelope including a second plurality of formant peaks corresponding to the first plurality of formant peaks, the second plurality of formant peaks having approximately equal amplitudes; and

(e) deriving a final set of filter coefficients from the second filtered DS signal.

26. The method of claim 25, wherein step (e) comprises:

(e)(i) performing an LPC analysis on the second filtered DS signal, to derive a third set of filter coefficients;

(e)(ii) bandwidth expanding the third set of filter coefficients to produce a fourth set of filter coefficients; and

(e)(iii) smoothing the fourth set of filter coefficients to produce the final set of filter coefficients.

27. A computer program product (CPP) comprising a computer usable medium having computer readable program code (CRPC) means embodied in the medium for causing an application program to execute on a computer processor to perform processing of a decoded speech (DS) signal, the DS signal having a spectral envelope including a first plurality of formant peaks, the CRPC means comprising:

first CRPC means for causing the processor to produce, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened time-domain DS signal having a spectral envelope including a second plurality of formant peaks corresponding to the first plurality o formant peaks, wherein one or more amplitude differences between respective ones of the second plurality of formant peaks are less than one or more corresponding amplitude differences between respective ones of the first plurality of formant peaks; and

second CRPC means for causing the processor to derive a set of filter coefficients a, from the spectrally-flattened time-domain DS signal.

28. The CPP of claim 27, wherein the first CRPC means includes:

third CRPC means for causing the processor to filter the DS signal to produce an intermediate spectrally-flattened DS signal having a spectral envelope including an intermediate plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the intermediate plurality of formant peaks are less than the one or more amplitude differences between the respective ones of the first plurality of formant peaks; and

fourth CRPC means for causing the processor to filter the intermediate spectrally-flattened DS signal to reduce the one or more amplitude differences between the respective ones of the intermediate plurality of formant peaks, thereby producing the spectrally-flattened DS signal.

29. The CPP of claim 28, wherein the first CRPC means further includes:

fifth CRPC means for causing the processor to derive an intermediate set of filter coefficients based on the DS signal,

wherein the third CRPC means includes CRPC means for causing the processor to filter the DS signal using the intermediate set of filter coefficients.

30. The CPP of claim 29, wherein the first CRPC means further includes:

sixth CRPC means for causing the processor to derive a second intermediate set of filter coefficients from the intermediate DS signal,

wherein the fourth CRPC means includes CRPC means for causing the processor to filter the intermediate spectrally-flattened DS signal using the second intermediate set of filter coefficients.

31. The CPP of claim 27, further comprising:

third CRPC means for causing the processor to bandwidth expand the set of filter coefficients to produce a set of bandwidth expanded filter coefficients; and

fourth CRPC means for causing the processor to smooth the bandwidth expanded set of filter coefficients to produce a smoothed set of filter coefficients.

32. The CPP of claim 31, further comprising:

fifth CRPC means for causing the processor to filter the DS signal using the smoothed set of filter coefficients.

33. The CPP of claim 27, wherein the set of filter coefficients are one or a set of all-pole filter coefficients, a set of pole-zero filter coefficients, and a set of all-zero filter coefficients.

34. An apparatus for processing a decoded speech (DS) signal, the DS signal having a spectral envelope including a first plurality of formant peaks, comprising:

first means for producing, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened time-domain DS signal having a spectral envelope including a second plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the second plurality of formant peaks are less than one or more corresponding amplitude differences between respective ones of the first plurality of formant peaks; and

second means for deriving a set of filter coefficients from the spectrally-flattened time-domain DS signal.

35. The apparatus of claim 34, wherein the first means comprises:

means for filtering the DS signal to produce an intermediate spectrally-flattened DS signal having a spectral envelope including an intermediate plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the intermediate plurality of formant peaks are less than the one or more amplitude differences between the respective ones of the first plurality of formant peaks; and

means for filtering the intermediate spectrally-flattened DS signal to reduce the one or more amplitude differences between the respective ones of the intermediate plurality of formant peaks, thereby producing the spectrally-flattened DS signal.

36. An apparatus for processing a decoded speech signal, the DS signal including a plurality of formant peaks, comprising:

a filter controller including

a first controller stage configured to produce, from the DS signal, a spectrally-flattened DS signal that is a time-domain signal, the spectrally-flattened time-domain DS signal having a spectral envelope including a second plurality of formant peaks corresponding to the first plurality of formant peaks, wherein one or more amplitude differences between respective ones of the second plurality of formant peaks are less than one or more corresponding amplitude differences between respective ones of the first plurality of formant peaks, and

a second controller stage configured to derive a set of filter coefficients a_t from the spectrally-flattened time-domain DS signal.

37. The apparatus of claim 36, further comprising:

a filter for filtering the DS signal, the filter being configured to receive the set of filter coefficients and having a frequency response controlled in accordance with the received set of filter coefficients.