Cross-talk cancellation

1. A method for cross-talk cancellation, the method comprising:

generating a cross-talk estimate signal using at least one hardware processor, the generating comprising:

delaying a far-end acoustic signal;

filtering the delayed far-end acoustic signal to produce a filter output, the filtering based at least in part on fixed predetermined filter coefficients, the delayed far-end acoustic signal being in the time domain;

adapting a gain variable for estimating cross-talk, the adapting the gain variable being a function of a cross-correlation and a magnitude of filter outputs, the cross-correlation being based at least in part on a near-end acoustic signal and the filter output; and

applying a gain to the filter output to produce the cross-talk estimate signal, the gain based at least in part on the adapted gain variable; and

subtracting the cross-talk estimate signal from the near-end acoustic signal to generate a cleaned acoustic signal.

2. The method of claim 1 , wherein the near-end acoustic signal is received via a microphone and the far-end acoustic signal is received via a receiver.

3. The method of claim 1 , wherein delaying the far-end acoustic signal is by a predetermined delay; and

wherein the cross-talk estimate signal is a scaled version of the filter outputs, the filter outputs being scaled using the adapted gain variable.

4. The method of claim 1 , wherein the adapting the gain variable comprises dividing the cross-correlation by the magnitude of the filter outputs.

5. The method of claim 1 , wherein the adapting the gain variable further comprises:

calculating an echo return loss enhancement (ERLE) based at least in part on a cross-correlation, the near-end acoustic signal, and the filter output; and

calculating the gain variable using the cross-correlation and the filter output if the ERLE is greater than a predetermined minimum threshold established for gain adaptation and the cross-correlation is greater than zero.

6. The method of claim 5 , wherein the cross-correlation is a cross product of a vector of the near-end acoustic signal and a vector of the filter output.

7. The method of claim 5 , wherein the ERLE is a magnitude squared of the near-end acoustic signal multiplied by an inverse of the cross-correlation squared divided by a magnitude squared of the filter output, subtracted from a magnitude squared of the near-end acoustic signal.

8. The method of claim 5 , wherein the gain variable is equal to the cross-correlation divided by a magnitude squared of the filter output.

9. The method of claim 5 , further comprising preventing divergence of the gain variable by:

placing a predetermined upper bound on the gain variable;

setting a state variable to a predetermined non-zero value in response to adapting the gain variable; and

providing the cleaned acoustic signal to an output instead of the near-end acoustic signal if the state variable has a non-zero value;

wherein the state variable is set to zero in response to a magnitude squared of a cleaned acoustic signal vector subtracted from a magnitude squared of a near-end acoustic signal vector being less than a threshold value.

10. A system for reducing cross-talk, the system comprising:

a delay element to delay a far-end acoustic signal by a predetermined delay;

a filter to receive the delayed far-end acoustic signal and generate a filter output based on fixed predetermined filter coefficients, the delayed far-end acoustic signal being in the time domain;

a gain control module, executable using a hardware processor, to apply an adapted gain to the filter output to produce a cross-talk estimate signal, the adapted gain being based at least in part on dividing a cross-correlation by a magnitude of filter outputs, the cross-correlation being based at least in part on a near-end acoustic signal and the filter output, the cross-talk estimate signal being a scaled filter output;

an adder configured to subtract the cross-talk estimate signal from the near-end acoustic signal to generate a cleaned acoustic signal; and

a switch to selectively output the near-end acoustic signal or the cleaned acoustic signal.

11. The system of claim 10 , wherein the filter comprises an n-tap finite impulse response (FIR) filter.

12. The system of claim 10 , wherein the near-end acoustic signal is received by a microphone and the far-end acoustic signal is received by a receiver.

13. A non-transitory computer-readable storage medium having embodied thereon a program, the program being executable by a processor to perform a method for cross-talk cancellation, the method comprising:

generating a cross-talk estimate signal using at least one hardware processor, the generating comprising:

delaying a far-end acoustic signal;

filtering the delayed far-end acoustic signal to produce a filter output, the filtering based at least in part on fixed predetermined filter coefficients, the delayed far-end acoustic signal being in the time domain;

adapting a gain variable for estimating cross-talk, the adapting the gain variable being a function of a cross-correlation and a magnitude of filter outputs, the cross-correlation being based at least in part on a near-end acoustic signal and the filter output; and

applying a gain to the filter output to produce the cross-talk estimate signal, the gain based at least in part on the adapted gain variable; and

subtracting the cross-talk estimate signal from the near-end acoustic signal to generate a cleaned acoustic signal.

14. The non-transitory computer-readable storage medium of claim 13 , wherein the adapting the gain variable further comprises:

calculating an echo return loss enhancement (ERLE) based at least in part on a cross-correlation, the near-end acoustic signal, and the filter output; and

calculating the gain variable using the cross-correlation and the filter output if the ERLE is greater than a predetermined minimum threshold established for gain adaptation and the cross-correlation is greater than zero.

15. The non-transitory computer-readable storage medium of claim 14 , wherein the cross-correlation is a cross product of a vector of the near-end acoustic signal and a vector of the filter output.

16. The non-transitory computer-readable storage medium of claim 14 , wherein the ERLE is a magnitude squared of the near-end acoustic signal multiplied by an inverse of the cross-correlation squared divided by a magnitude squared of the filter output, subtracted from a magnitude squared of the near-end acoustic signal.

17. The non-transitory computer-readable storage medium of claim 14 , wherein the gain variable is equal to the cross-correlation divided by a magnitude squared of the filter output.

18. The non-transitory computer-readable storage medium of claim 14 , the method further comprising preventing divergence of the gain variable by:

placing a predetermined upper bound on the gain variable;

setting a state variable to a predetermined non-zero value in response to adapting the gain variable; and

providing the cleaned acoustic signal to an output instead of the near-end acoustic signal if the state variable has a non-zero value;

wherein the state variable is set to zero in response to a magnitude squared of a cleaned acoustic signal vector subtracted from a magnitude squared of a near-end acoustic signal vector being less than a threshold value.

19. The non-transitory computer-readable storage medium of claim 13 , wherein the adapting the gain variable comprises dividing the cross-correlation by the magnitude of the filter outputs.

20. The non-transitory computer-readable storage medium of claim 13 , wherein the near-end acoustic signal is received via a microphone and the far-end acoustic signal is received via a receiver.