Microphone array signal enhancement

• US 20030206640A1
• Filed: 05/02/2002
• Published: 11/06/2003
• Est. Priority Date: 05/02/2002
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Malvar, Henrique S., Florencio, Dinei Afonso Ferreira, Gillespie, Bradford W.
• Granted Patent: US 7,167,568 B2
• Time in Patent Office: Days
• US Class Current: 381/93
• CPC Class Codes: H03H 21/0012 Digital adaptive filters
• CPC Class Codes: H04R 1/005 using digitally weighted transducing elements

Abstract

A system and method facilitating signal enhancement utilizing an adaptive filter is provided. The invention includes an adaptive filter that filters an input based upon a plurality of adaptive coefficients, the adaptive filter modifying at least one of the adaptive coefficients based on a feedback output. The invention further includes a feedback component that provides the feedback output based, at least in part, upon a non-linear function of the acoustic reverberation reduced output.

The invention further provides a noise statistics component that stores noise statistics associated with a noise portion of an input signal and a signal+noise statistics component that stores signal+noise statistics associated with a signal and noise portion of the input signal. The invention further provides a spatial filter that provides an output signal based, at least in part, upon a filtered input signal, the filtering being based, at least in part, upon a weighted error calculation of the noise statistics and the signal+noise statistics.

Claims

1. An audio enhancement system, comprising:

an adaptive filter that filters an input signal, at least in part, upon a plurality of adaptive coefficients, the adaptive filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output, the adaptive filter providing a quality enhanced output; and

,

a feedback component that provides the feedback output based, at least in part, upon a non-linear function of the quality enhanced output.

2. The audio enhancement system of claim 1, the non-linear function being based on increasing a non-linear measure of speechness.

3. The audio enhancement system of claim 2, the non-linear measure of speechness being based on maximizing kurtosis of a LPC residual of the quality enhanced output.

4. The audio enhancement system of claim 2, the non-linear measure of speechness being based on maximizing the impulse-train like characteristic of a LPC residual of the quality enhanced output.

5. The audio enhancement system of claim 1, the input signal being based on one microphone.

6. The audio enhancement system of claim 1, the input signal being based on at least two microphones.

7. An audio enhancement system, comprising:

a first adaptive filter that filters an input signal based, at least in part, upon a plurality of adaptive coefficients, the first adaptive filter providing a quality enhanced output;

a second adaptive filter that filters the input signal based, at least in part, upon the plurality of adaptive coefficients, the second adaptive filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output, the second adaptive filter further providing an output; and

,

a feedback component that provides the feedback output based, at least in part, upon a non-linear function of the output of the second adaptive filter.

8. The audio enhancement system of claim 7, the second adaptive filter being adapted based on a delayed version of the input signal.

9. The audio enhancement system of claim 7, the second adaptive filter being adapted based on a modified version of the input signal.

10. The audio enhancement system of claim 7, the second adaptive filter being adapted based on an equalized version of the input signal.

11. The audio enhancement system of claim 7, the second adaptive filter running at a different data rate than the first adaptive filter.

13. The audio enhancement system of claim 11, the second adaptive filter running at a higher data rate than the first adaptive filter.

14. The audio enhancement system of claim 7, the second adaptive filter using data stored in the system.

15. The audio enhancement system of claim 14, the second adaptive filter using the data stored in the system more than once.

16. The audio enhancement system of claim 14, the second adaptive filter using the data stored in the system in a different order than the time-of-arrival order.

17. The audio enhancement system of claim 7, the adaptive filter using data modified to edit out non-voiced parts of the input signal.

18. The audio enhancement system of claim 7, further comprising a plurality of second adaptive filters.

19. The audio enhancement system of claim 18, the first adaptive filter and the plurality of second adaptive filters employing the same plurality of adaptive coefficients, the plurality of adaptive coefficients being modified, at least in part, based on an output of at least one of the second adaptive filters.

20. The audio enhancement system of claim 19, adaptation feedback provided by at least one of the second adaptive filters being based on a non-linear function.

12. The audio enhancement system of claim 111, the second adaptive filter running at a lower data rate than the first adaptive filter.

21. An acoustic reverberation reduction system, comprising:

an adaptive filter that filters a linear prediction output based, at least in part, upon a plurality of adaptive coefficients, the adaptive filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output, the adaptive filter providing an acoustic reverberation reduced output; and

,

a feedback component that provides the feedback output based, at least in part, upon a non-linear function of the acoustic reverberation reduced output.

22. The acoustic reverberation system of claim 21, further comprising a linear prediction analyzer that analyzes an input signal and provides the linear prediction residual output.

23. The acoustic reverberation system of claim 22, further comprising a linear prediction synthesis filter that filters the acoustic reverberation output and provides a processed output signal.

24. The acoustic reverberation reduction system of claim 21, the non-linear function being based, at least in part, upon maximization of kurtosis.

25. The acoustic reverberation reduction system of claim 24, the non-linear function being based, at least in part, upon the following equation:

J(n)=E{{tilde over (y)}⁴(n)}/E²{{tilde over (y)}²(n)}−

3

where J(n) is the kurtosis metric,

y(n) is the acoustic reverberation reduced output, and,

and E{ } are expectations.

26. The acoustic reverberation reduction system of claim 24, the non-linear function being based, at least in part, upon selecting a threshold, and providing positive feedback for samples above that threshold and negative feedback for samples below that threshold.

27. The acoustic reverberation system of claim 21, the feedback component employing LPC filtering of the acoustic reverberation reduced output, the feedback output being based, at least in part, upon a non-linear function based on a LPC residual.

28. The acoustic reverberation system of claim 21, the feedback component employing LPC filtering of the acoustic reverberation reduced output, the feedback output being based, at least in part, upon a non-linear function based on a LPC residual which increases the impulse-train characteristics of the LPC residual.

29. The acoustic reverberation system of claim 21, the feedback component employing LPC filtering of the acoustic reverberation reduced output, the feedback output being based, at least in part, upon a non-linear function based on a LPC residual which increases the kurtosis of the LPC residual.

30. An acoustic reverberation reduction system, comprising:

a first adaptive filter that filters an input signal based, at least in part, upon a plurality of adaptive coefficients, the first adaptive filter providing an acoustic reverberation reduced output;

a second adaptive filter that filters a linear prediction output based, at least in part, upon the plurality of adaptive coefficients, the second adaptive filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output, the second adaptive filter further providing an output; and

,

a feedback component that provides the feedback output based, at least in part, upon a non-linear function of the output of the second adaptive filter.

31. The acoustic reverberation reduction system of claim 30, further comprising a linear prediction analyzer that analyzes the input signal and provides the linear prediction residual output;

32. The acoustic reverberation reduction system of claim 30, the non-linear function being based, at least in part, upon maximization of kurtosis.

33. The acoustic reverberation reduction system of claim 32, the non-linear function being based, at least in part, upon the following equation:

J(n)=E{{tilde over (y)}⁴(n)}/E²{{tilde over (y)}²(n)}−

3

where J(n) is the kurtosis metric,

y(n) is the acoustic reverberation reduced output, and,

and E{ } are expectations.

34. The acoustic reverberation reduction system of claim 30, the first adaptive filter further comprising a frequency transform component that performs a frequency domain transform of the input signal, filtering of the first adaptive filter being performed on the frequency transformed input signal, the first adaptive filter further comprising an inverse frequency transform component that performs an inverse frequency domain transform of the filtered frequency transformed input signal.

35. The acoustic reverberation reduction system of claim 34, the frequency domain transform being a Modulated Complex Lapped Transform.

36. The acoustic reverberation reduction system of claim 34, the second adaptive filter further comprising a frequency transform component that performs a frequency domain transform of the input signal, filtering of the second adaptive filter being performed on the frequency transformed input signal, the second adaptive filter further comprising an inverse frequency transform component that performs an inverse frequency domain transform of the filtered frequency transformed input signal.

37. The acoustic reverberation reduction system of claim 36, the frequency domain transform of the second adaptive filter being a Modulated Complex Lapped Transform.

38. An acoustic reverberation reduction system, comprising:

a plurality of reverberation reduction channel components, at least some of the reverberation reduction channel components comprising a first adaptive filter comprising a frequency transform component that performs a frequency domain transform of an input signal of the channel, the first adaptive filter filtering the frequency transformed input signal based, at least in part, upon a plurality of adaptive coefficients, the first adaptive filter further comprising an inverse frequency transform component that performs an inverse frequency domain transform of the filtered frequency transformed input signal and provides an acoustic reverberation reduced output;

a linear prediction analyzer that analyzes the input signal of the channel and provides a linear prediction residual output;

a second adaptive filter comprising a frequency transform component that performs a frequency domain transform of the linear prediction residual output, the second adaptive filter filtering the linear prediction residual output based, at least in part, upon a plurality of adaptive coefficients, the second adaptive filter further comprising an inverse frequency transform component that performs an inverse frequency domain transform of the filtered frequency transformed linear prediction residual output and provides an output; and

a summing component that sums the plurality of outputs of the second adaptive filters; and

,

a feedback component that modifies at least one of the plurality of adaptive coefficients based, at least in part, upon a non-linear function of the output of summing component.

39. The acoustic reverberation reduction system of claim 38, the non-linear function being based, at least in part, upon maximization of kurtosis.

40. The acoustic reverberation reduction system of claim 38, the frequency domain transform of at least one of the first adaptive filter and the second adaptive filter being a Modulated Complex Lapped Transform.

41. The acoustic reverberation reduction system of claim 38, data used to drive the second adaptive filter being stored in the system.

42. The acoustic reverberation reduction system of claim 41, the data used to drive the second adaptive filter not being synchronized with the input signal.

43. The acoustic reverberation reduction system of claim 38, the data used to drive the second adaptive filter being based, at least in part, upon the presence of voiced speech.

44. The acoustic reverberation reduction system of claim 38, a data rate of the first and second adaptive filters being different.

45. The acoustic reverberation reduction system of claim 38, the data on each channel corresponding to a signal from different microphones.

46. An acoustic noise reduction system, comprising:

a first signal statistics component that stores statistics associated with a certain portion or type of an input signal;

a second signal statistics component that stores statistics associated with a second portion or type of the input signal; and

,

a spatial filter that provides an output signal, the output signal being based, at least in part, upon a filtered input signal, the filtering being based, at least in part, upon a weighted error calculation based on the two classes of statistics.

47. The acoustic noise reduction system of claim 46, the first signal statistics component being the statistics of noise-like segments of the input signal, and, the second signal statistics component being the statistics of signal+noise segments of the input signal.

48. The acoustic noise reduction system of claim 46, further comprising a voice activity detector that facilitates storing of at least one of the first statistics and the second statistics.

49. The acoustic noise reduction system of claim 46, filtering being performed in the frequency domain.

50. An acoustic noise reduction system, comprising:

a filter that filters an input signal based, at least in part, upon a plurality of adaptive coefficients, the first adaptive filter providing an acoustic noise reduced output;

a signal+noise buffer that stores a signal+noise portion of the input signal;

a noise buffer that stores a noise portion of the input signal;

a signal composer that generates a synthetic input signal and a synthetic desired signal based, at least in part, upon information stored in at least one of the signal+noise buffer and the noise buffer

a LMS filter that filters the synthetic input signal based, at least in part, upon the plurality of adaptive coefficients, the filtering being based, at least in part, upon a weighted error calculation of noise statistics and signal+noise statistics, the LMS filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output, the LMS filter further providing an output; and

,

a differential component that provides the feedback output based, at least in part, upon a difference between the LMS filter output and the synthetic desired signal.

51. The acoustic noise reduction system of claim 50, the filtering of at least one of the first adaptive filter and the second adaptive filter being done in the frequency domain.

52. An acoustic signal enhancement system, comprising:

a voice activity noise detector that provides information to a noise buffer and a signal+noise buffer;

a filter that filters the input signal based, at least in part, upon a plurality of adaptive coefficients, the filter providing an acoustic enhanced signal output;

a noise adaptive filter that that filters information stored in the noise buffer based, at least in part, upon the plurality of adaptive coefficients, the noise reduction adaptive filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a noise reduction feedback output, the noise reduction adaptive filter further providing an output;

a noise feedback component that provides the noise reduction feedback output based, at least in part, upon a weighted error calculation of the output of the noise reduction adaptive filter.

a reverberation adaptive filter that filters at least a part of the speech signal based, at least in part, upon the plurality of adaptive coefficients, the reverberation reduction adaptive filter modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a reverberation feedback output, the reverberation reduction adaptive filter further providing an output; and

,

a reverberation feedback component that provides the reverberation feedback output based, at least in part, upon a non-linear function of the output of the reverberation reduction adaptive filter.

53. A method for reducing acoustic reverberation, comprising:

providing an acoustic reverberation reduced output based, at least in part, upon a plurality of adaptive coefficients;

modifying at least one of the plurality of adaptive coefficient based, at least in part, upon a feedback output; and

,

providing the feedback output based, at least in part, upon a non-linear function of the acoustic reverberation reduced output.

54. The method of claim 53, the non-linear function being based, at least in part, upon maximization of kurtosis.

55. A method for reducing acoustic reverberation, comprising:

filtering an input signal based, at least in part, upon a plurality of adaptive coefficients;

providing an acoustic reverberation reduced output based, at least in part, upon the filtered input signal;

providing a linear prediction residual output based on the input signal;

filtering the linear prediction residual output based, at least in part, upon the plurality of adaptive coefficients;

modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output; and

,

providing the feedback output based, at least in part, upon a non-linear function of the filtered linear prediction residual output.

56. The method of claim 55, the non-linear function being based, at least in part, upon maximization of kurtosis.

57. A method for reducing acoustic noise, comprising:

storing noise statistics associated with a noise portion of an input signal;

storing signal+noise statistics with a signal and noise portion of the input signal; and

,

filtering the input signal based, at least in part, upon a weighted error calculation of the noise statistics and the signal+noise statistics.

58. A method for reducing acoustic noise, comprising:

storing a noise portion of an input signal;

storing a signal+noise portion of the input signal;

filtering the input signal based, at least in part, upon a plurality of adaptive coefficients;

generating a synthetic input signal and a synthetic desired signal based, at least in part, upon at least one of the stored noise portion of the input signal and the stored signal+noise portion of the input signal;

providing an LMS filter output based, at least in part, upon the plurality of adaptive coefficients, the filtering being based, at least in part, upon a weighted error calculation of noise statistics and signal+noise statistics,

modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output; and

,

providing the feedback output based, at least in part, upon a difference between the LMS filter output and the synthetic desired signal.

59. A method for enhancing an acoustic signal, comprising:

storing noise information associated with a noise portion of an input signal;

storing signal+noise information with a signal and noise portion of the input signal;

filtering the input signal based, at least in part, upon a plurality of adaptive coefficients;

providing a noise adaptive filter output signal based, at least in part, upon the stored noise information and the plurality of adaptive coefficients;

modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a noise feedback output;

providing the noise feedback output based, at least in part, upon a weighted error of the noise statistics and the signal+noise statistics;

providing a reverberation filter output based, at least in part, upon the stored signal+noise information and the plurality of adaptive coefficients;

modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a reverberation feedback output; and

,

providing the reverberation feedback output based, at least in part, upon a non-linear function of the output of the reverberation adaptive filter.

60. A data packet transmitted between two or more computer components that facilitates acoustic reverberation reduction, the data packet comprising:

a data field comprising a plurality of adaptive coefficients, at least one of the plurality of adaptive coefficients having been modified based, at least in part, upon a feedback output based, at least in part, upon a non-linear function of an acoustic reverberation reduced output.

61. A data packet transmitted between two or more computer components that facilitates acoustic noise reduction, the data packet comprising:

a data field comprising a plurality of adaptive coefficients, at least one of the plurality of adaptive coefficients having been modified based, at least in part, upon a feedback output based, at least in part, upon a weighted error calculation of noise statistics and signal+noise statistics.

62. A computer readable medium storing computer executable components of a system facilitating acoustic reverberation reduction, comprising:

an adaptive filter component that filters a linear prediction output based, at least in part, upon a plurality of adaptive coefficients, the adaptive filter component modifying at least one of the plurality of adaptive coefficients based, at least in part, upon a feedback output, the adaptive filter component providing an acoustic reverberation reduced output; and

,

a feedback component that provides the feedback output based, at least in part, upon a non-linear function of the acoustic reverberation reduced output.

63. A computer readable medium storing computer executable components of a system facilitating acoustic noise reduction, comprising:

a noise statistics component that stores noise statistics associated with a noise portion of an input signal;

a signal+noise statistics component that stores signal+noise statistics associated with a signal and noise portion of the input signal; and

,

a spatial filter component that provides an output signal, the output signal being based, at least in part, upon a filtered input signal, the filtering being based, at least in part, upon a weighted error calculation of the noise statistics and the signal+noise statistics.

64. An acoustic reverberation reduction system, comprising:

means for providing an acoustic reverberation reduced output based, at least in part, upon a plurality of adaptive coefficients;

means for modifying at least one of the plurality of adaptive coefficient based, at least in part, upon a feedback output; and

,

means for providing the feedback output based, at least in part, upon a non-linear function of the acoustic reverberation reduced output.

65. An acoustic noise reduction system, comprising:

means for storing noise statistics associated with a noise portion of an input signal;

means for storing signal+noise statistics with a signal and noise portion of the input signal; and

,

means for filtering the input signal based, at least in part, upon a weighted error calculation of the noise statistics and the signal+noise statistics.