Method and system for clear signal capture

• US 20070041575A1
• Filed: 08/10/2005
• Published: 02/22/2007
• Est. Priority Date: 08/10/2005
• Status: Active Grant

• Current Assignee: CSR Technology Incorporated (Qualcomm, Inc.)
• Original Assignee: Clarity Technologies, Inc. (Qualcomm, Inc.)
• Inventors: Yen, Kuan-Chieh, Huat Ngia, Lester, Alves, Rogerio
• Granted Patent: US 7,426,270 B2
• Time in Patent Office: Days
• US Class Current: 379/406.80
• CPC Class Codes: G10L 2021/02082 the noise being echo, reverberation of the speech
• CPC Class Codes: H04M 9/082 using echo cancellers echo cancellers per se H04B3/23

Abstract

A method and system for clear signal capture comprehend several individual aspects that address specific problems in improved ways. In addition, the method and system also comprehend a hands-free implementation that is a practical solution to a very complex problem.

Claims

1. A method of acoustic echo cancellation (AEC) wherein a microphone signal resulting from an unobservable signal corrupted by both additive background noise and an acoustic echo is processed in an attempt to restore the unobservable signal, the acoustic echo being a speaker signal modified by an acoustic path, wherein an adaptive filter models the echo path, the method comprising:

receiving the microphone signal;

applying a first asymmetric nonlinear adaptive filter in fullband to the speaker signal to produce a first output signal;

applying a first adaptive filter in fullband to the first output signal to produce a first echo signal that models the acoustic echo;

subtracting the first echo signal from the microphone signal to produce a first echoless signal that resembles the unobservable signal; and

adapting the first nonlinear adaptive filter in fullband and the first adaptive filter based on the first echoless signal.

2. The method of claim 1 wherein the first asymmetric nonlinear adaptive filter is composed of a cascade of at least two nonlinear systems including a first system representing an upper clipping point and a second system representing a lower clipping point.

3. The method of claim 1 further comprising:

applying a double talk control, obtained from a frequency domain implementation, to modify an update equation of clipping points of the first asymmetric nonlinear adaptive filter.

4. The method of claim 1 further comprising:

selectively applying a second asymmetric nonlinear adaptive filter in fullband, to the speaker signal to produce a second output signal, the second asymmetric nonlinear adaptive filter being set based on the first asymmetric nonlinear adaptive filter;

applying a main adaptive filter in fullband to the second output signal to produce a second echo signal that models the acoustic echo;

subtracting the second echo signal from the microphone signal in frequency domain to produce a second echoless signal that resembles the unobservable signal; and

adapting the main adaptive filter in frequency domain based on the second echoless signal.

5. The method of claim 4 wherein the first asymmetric nonlinear adaptive filter is composed of a cascade of at least two nonlinear systems including a first system representing an upper clipping point and a second system representing a lower clipping point.

6. The method of claim 4 further comprising:

applying a double talk control, obtained from a frequency domain implementation, to modify an update equation of clipping points of the first asymmetric nonlinear adaptive filter.

7. The method of claim 4 wherein the selective application of the second asymmetric nonlinear adaptive filter is based on a comparison of the speaker signal to clipping points defined by the first asymmetric nonlinear adaptive filter.

8. The method of claim 7 wherein the second asymmetric nonlinear adaptive filter is disabled when the clipping points are smaller than a predetermined threshold.

9. The method of claim 7 wherein the second asymmetric nonlinear adaptive filter is disabled when the echo return loss enhancement (ERLE) degrades when the clipping points are applied.

10. A method of acoustic echo cancellation (AEC) wherein a microphone signal resulting from an unobservable signal corrupted by both additive background noise and an acoustic echo is processed in an attempt to restore the unobservable signal, the acoustic echo being a speaker signal modified by an acoustic path, wherein an adaptive filter in frequency domain models the echo path, the method including receiving the microphone signal, applying an adaptive filter in subbands to the speaker signal to produce an echo signal that models the acoustic echo, subtracting the echo signal from the microphone signal to produce a frequency domain echoless signal that resembles the unobservable signal, and adapting the adaptive filter based on the frequency domain echoless signal, the method further comprising:

applying a residual echo reduction filter to the frequency domain echoless signal;

determining an average bandwidth attenuation factor based on the coefficients of the residual echo reduction filter;

determining a threshold; and

when the average bandwidth attenuation factor is less than the threshold, applying a frequency dependent nonlinear attenuation vector to a defined group of frequency bands.

11. The method of claim 10 wherein the defined group of frequency bands is a group of high frequency bands, thereby providing frequency dependent nonlinear processing to further reduce high frequency echo.

12. A method of acoustic echo cancellation (AEC) wherein a microphone signal resulting from an unobservable signal corrupted by both additive background noise and an acoustic echo is processed in an attempt to restore the unobservable signal, the acoustic echo being a speaker signal modified by an acoustic path, wherein an adaptive filter in frequency domain models the echo path, the method including receiving the microphone signal, applying an adaptive filter in subbands to the speaker signal to produce an echo signal that models the acoustic echo, subtracting the echo signal from the microphone signal to produce a frequency domain echoless signal that resembles the unobservable signal, and adapting the adaptive filter based on the frequency domain echoless signal, the method further comprising:

making a classification of near-end and far-end speeches to detect a significant nonlinear relationship between near-end and far-end speeches; and

in the event that a significant nonlinear relationship is detected, increasing the adaptation rate of the adaptive filter in frequency domain to escape local minima and approach the global minima.

13. The method of claim 12 wherein making the classification involves considering the coherence of the microphone signal and the echo signal produced by the adaptive filter in frequency domain.

14. The method of claim 12 wherein the adaptive filter in frequency domain is composed of a main filter and an auxiliary filter, the auxiliary filter being utilized in certain predetermined situations during significant nonlinear relationship that causes the main filter underestimates echo, wherein increasing the adaptation rate involves increasing the adaptation rate of the auxiliary filter to result in an increase of the adaptation rate of main filter.

15. A method of acoustic echo cancellation (AEC) wherein a microphone signal resulting from an unobservable signal corrupted by both additive background noise and an acoustic echo is processed in an attempt to restore the unobservable signal, the acoustic echo being a speaker signal modified by an acoustic path, wherein an adaptive filter in frequency domain models the echo path, the method including receiving the microphone signal, applying an adaptive filter in subbands to the speaker signal to produce an echo signal that models the acoustic echo, subtracting the echo signal from the microphone signal to produce a frequency domain echoless signal that resembles the unobservable signal, and adapting the adaptive filter based on the frequency domain echoless signal, the method further comprising:

applying a residual echo reduction filter to the frequency domain echoless signal;

utilizing state differentiators to make a classification of near-end and far-end speeches to detect a significant nonlinear relationship between near-end and far-end speeches; and

in the event that a significant nonlinear relationship is detected, modifying the residual echo reduction filter to provide additional attenuation to the residual echo relative to the extent of attenuation provided in the absence of significant nonlinearity.

16. The method of claim 15 wherein the classification involves considering the coherence of the microphone signal and the echo signal produced by the adaptive filter in frequency domain.

17. The method of claim 16 wherein a gain constant applied to the residual echo reduction filter reflects an average gain that has a relationship to the coherence of the microphone signal and the echo signal produced by the adaptive filter in frequency domain.

18. The method of claim 16 wherein the classification further involves considering the expected coherence of the microphone signal and the frequency domain echoless signal.

19. The method of claim 18 wherein the classification is based on a difference between average values of the microphone signal/echo signal coherence and the microphone signal/frequency domain echoless signal expected coherence.

20. The method of claim 19 further comprising:

determining a threshold; and

when the determined difference between average values is less than the threshold, indicating significant nonlinearity, applying a frequency dependent nonlinear attenuation vector to a defined group of frequency bands.

21. A method of acoustic echo cancellation (AEC) wherein a microphone signal resulting from an unobservable signal corrupted by both additive background noise and an acoustic echo is processed in an attempt to restore the unobservable signal, the acoustic echo being a speaker signal modified by an acoustic path, wherein an adaptive filter models the echo path, the method including receiving the microphone signal, applying an adaptive filter to the speaker signal to produce an echo signal that models the acoustic echo, subtracting the echo signal from the microphone signal to produce a first echoless signal that resembles the unobservable signal, and adapting the adaptive filter based on the first echoless signal, the method further comprising:

prior to the input of the adaptive filter, a nonlinear compensation scaled by volume information of the system, is applied to the input signal to produce a new input of the adaptive filter and the speaker signal, whereby this dynamic nonlinear pre-distoration allows the system to be treated as a linear system.

22. The method of claim 21 wherein volume adjustments are made prior to applying the nonlinear compensation.

23. The method of claim 22 wherein the applied nonlinear compensation is a hard clip.

24. The method of claim 22 wherein the applied nonlinear compensation is a soft clip.