Method and system for double-end talk detection, and method and system for echo elimination

1. A method for eliminating echo in a speaker-microphone communication system, comprising

performing a noise energy estimating process on a local microphone signal in order to obtain an estimated noise signal, wherein the local microphone signal comprises a local voice signal, possible background noise signal, and possible remote voice signal output from a speaker and received by a microphone;

performing an echo estimating process on a remote voice signal to obtain an estimated echo signal, wherein the remote voice signal is output from the speaker;

determining an error signal from the local microphone signal and the estimated echo signal;

calculating a variance (σ e 2 ) of the error signal and a variance ({circumflex over (σ)} n 2 ) of the estimated noise signal;

calculating a determinant (ξ), wherein the determinant (ξ) corresponds to the variance (σ e 2 ) of the error signal and variance ({circumflex over (σ)} n 2 ) of the estimated noise signal; and

comparing the determinant (ξ) and a preset threshold, wherein when the determinant (ξ) is lower than the preset threshold, it is determined that double-end talk has not occurred, otherwise, it is determined that double-end talk has occurred.

2. The method as claimed in claim 1 , wherein the determinant ξ=c·|σ e 2 −{circumflex over (σ)} n 2 |, and ‘c’ is a constant.

3. The method as claimed in claim 1 , wherein the determinant

0

≤

ξ

=

1

-



σ

^

n

2

σ

e

2



.

4. The method as claimed in claim 1 , further comprising:

calculating a variance ({circumflex over (σ)} y 2 ) of the estimated echo signal,

wherein the determinant

ξ

=



σ

e

2

-

σ

^

n

2

σ

^

y

2



.

5. The method as claimed in claim 4 , wherein the variance (σ e 2 ) of the error signal, the variance ({circumflex over (σ)} n 2 ) of the estimated noise signal, and/or the variance ({circumflex over (σ)} y 2 ) of the estimated echo signal are calculated for a whole frequency domain or for a low frequency domain.

6. The method as claimed in claim 5 , wherein the determinant (ξ) is calculated for each frame, and the error signal of a current frame, the estimated noise signal of the current frame, and the estimated echo signal of the current frame are transferred, using N point Fast Fourier Transfer, to corresponding frequency domain signals, respectively, and the variance (σ e 2 ) of t the error signal, variance ({circumflex over (σ)} n 2 ) of the estimated noise signal, or variance ({circumflex over (σ)} y 2 ) of the estimated echo signal are calculated according to the following equations:

σ

e

2

⁡

(

i

)

=

λσ

e

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

E

j

⁡

(

i

)

σ

^

n

2

⁡

(

i

)

=

λ

⁢

σ

^

n

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

Noise

j

⁡

(

i

)

σ

^

y

2

⁡

(

i

)

=

λ

⁢

σ

^

y

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

Y

^

j

⁡

(

i

)

,

wherein σ e 2 (i) is the variance of the error signal in the i th frame, {circumflex over (σ)} n 2 (i) is the variance of the estimated noise signal in the i th frame, {circumflex over (σ)} y 2 (i) is the variance of the estimated echo signal in the i th frame, E j (i) represents energy at frequency j of the frequency domain signal of the error signal in the i th frame, Noise j (i) represents energy at frequency j of the frequency domain signal of the noise signal in the i th frame, and Ŷ j (i) represents energy at frequency j of the frequency domain signal of the estimated echo signal in the i th frame, wherein 0.9<λ<1, λ is a real number, 0<P≦N/2−1, P is a positive integral, and the value of P is determined by the non-linearity of the communication system.

7. The method as claimed in claim 1 , wherein when it is determined that double-end talk has occurred, the echo estimating adaptation process of the echo estimating coefficient is disabled.

8. A system for eliminating echo in a speaker-microphone communication system, comprising:

a noise energy estimating device, performing a noise energy estimating process on a local microphone signal in order to obtain an estimated noise signal, wherein the local microphone signal comprises a local voice signal, possible background noise signal, and possible remote voice signal output from a speaker and received by a microphone;

an echo estimating device, performing an echo estimating process on a remote voice signal to obtain an estimated echo signal, wherein the remote voice signal is output from the speaker;

an error signal determining device, determining an error signal from the local microphone signal and the estimated echo signal;

a determinant calculator, calculating a determinant (ξ), wherein the determinant (ξ) corresponds to the variance (σ e 2 ) of the error signal and variance ({circumflex over (σ)} n 2 ) of the estimated noise signal; and

a double-end talk detecting device, comparing the determinant (ξ) and a preset threshold, wherein when the determinant (ξ) is lower than the preset threshold, it is determined that double-end talk has not occurred, otherwise, it is determined that double-end talk has occurred.

9. The system as claimed in claim 8 , wherein the determinant calculator calculates the determinant as ξ=c·|σ e 2 −{circumflex over (σ)} n 2 |, and ‘c’ is a constant.

10. The system as claimed in claim 8 , wherein the determinant calculator calculates the determinant as

0

≤

ξ

=

1

-



σ

^

n

2

σ

e

2



.

11. The system as claimed in claim 8 , wherein the determinant calculator further performs the step of:

calculating a variance ({circumflex over (σ)} y 2 ) of the estimated echo signal,

wherein the determinant

ξ

=



σ

e

2

-

σ

^

n

2

σ

^

y

2



.

12. The system as claimed in claim 11 , wherein the determinant calculator calculates variance (σ e 2 ) of the error signal, the variance ({circumflex over (σ)} n 2 ) of the estimated noise signal, and/or the variance ({circumflex over (σ)} y 2 ) of the estimated echo signal for a whole frequency domain or for a low frequency domain.

13. The system as claimed in claim 12 , wherein the determinant calculator calculates the determinant (ξ) for each frame, and transfers, using N point Fast Fourier Transfer, the error signal of a current frame, the estimated noise signal of the current frame, and the estimated echo signal of the current frame to corresponding frequency domain signals, respectively, and calculates the variance (σ e 2 ) of the error signal, variance ({circumflex over (σ)} n 2 ) of the estimated noise signal, or variance ({circumflex over (σ)} y 2 ) of the estimated echo signal according to the following equations:

σ

e

2

⁡

(

i

)

=

λσ

e

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

E

j

⁡

(

i

)

σ

^

n

2

⁡

(

i

)

=

λ

⁢

σ

^

n

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

Noise

j

⁡

(

i

)

σ

^

y

2

⁡

(

i

)

=

λ

⁢

σ

^

y

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

Y

^

j

⁡

(

i

)

,

wherein σ e 2 (i) is the variance of the error signal in the i th frame, {circumflex over (σ)} n 2 (i) is the variance of the estimated noise signal in the i th frame, {circumflex over (σ)} y 2 (i) is the variance of the estimated echo signal in the i th frame, E j (i) represents energy at frequency j of the frequency domain signal of the error signal in the i th frame, Noise j (i) represents energy at frequency j of the frequency domain signal of the noise signal in the i th frame, and Ŷ j (i) represents energy at frequency j of the frequency domain signal of the estimated echo signal in the i th frame, wherein 0.9<λ<1, λ is a real number, 0<P≦N/2−1, P is a positive integral, and the value of P is determined by the non-linearity of the communication system.

14. The system as claimed in claim 8 , wherein when the double-end talk detecting device determines that double-end talk has occurred, the echo estimating adaptation process of the echo estimating coefficient is disabled.

15. A circuit for eliminating echo in a speaker-microphone communication system, wherein the system comprises a noise energy estimating device, an echo estimating device, and an error signal determining device, comprising:

a determinant calculator, calculating a determinant (ξ), wherein the determinant (ξ) corresponds to the variance (σ e 2 ) of an error signal from the error signal determining device and variance ({circumflex over (σ)} n 2 ) of an estimated noise signal from the noise energy estimating device; and

a double-end talk detecting device, comparing the determinant (ξ) and a preset threshold, wherein when the determinant (ξ) is lower than the preset threshold, it is determined that double-end talk has not occurred, otherwise, it is determined that double-end talk has occurred.

16. The circuit of claim 15 , wherein the determinant calculator calculates the determinant as ξ=c·|σ e 2 −{circumflex over (σ)} n 2 |, and ‘c’ is a constant.

17. The circuit of claim 15 , wherein the determinant calculator calculates the determinant as

0

≤

ξ

=

1

-



σ

^

n

2

σ

e

2



.

18. The circuit of claim 15 , wherein the determinant calculator further performs the step of:

calculating a variance ({circumflex over (σ)} y 2 ) of an estimated echo signal from the echo estimating device,

wherein the determinant

ξ

=



σ

e

2

-

σ

^

n

2

σ

^

y

2



.

19. The circuit of claim 18 , wherein the determinant calculator calculates variance (σ e 2 ) of the error signal, the variance ({circumflex over (σ)} n 2 ) of the estimated noise signal, and/or the variance ({circumflex over (σ)} y 2 ) of the estimated echo signal for a whole frequency domain or for a low frequency domain, wherein the determinant calculator calculates the determinant (ξ) for each frame, and transfers, using N point Fast Fourier Transfer, the error signal of a current frame, the estimated noise signal of the current frame, and the estimated echo signal of the current frame to corresponding frequency domain signals, respectively, and calculates the variance (σ e 2 ) of the error signal, variance ({circumflex over (σ)} n 2 ) of the estimated noise signal, or variance ({circumflex over (σ)} y 2 ) of the estimated echo signal according to the following equations:

σ

e

2

⁡

(

i

)

=

λσ

e

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

E

j

⁡

(

i

)

σ

^

n

2

⁡

(

i

)

=

λ

⁢

σ

^

n

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

Noise

j

⁡

(

i

)

σ

^

y

2

⁡

(

i

)

=

λ

⁢

σ

^

y

2

⁡

(

i

-

1

)

+

(

1

-

λ

)

⁢

∑

j

=

0

P

⁢

Y

^

j

⁡

(

i

)

,

wherein σ e 2 (i) is the variance of the error signal in the i th frame, {circumflex over (σ)} n 2 (i) is the variance of the estimated noise signal in the i th frame, {circumflex over (σ)} y 2 (i) is the variance of the estimated echo signal in the i th frame, E j (i) represents energy at frequency j of the frequency domain signal of the error signal in the i th frame, Noise j (i) represents energy at frequency j of the frequency domain signal of the noise signal in the i th frame, and Ŷ j (i) represents energy at frequency j of the frequency domain signal of the estimated echo signal in the i th frame, wherein 0.9<λ<1, λ is a real number, 0<P≦N/2−1, P is a positive integral, and the value of P is determined by the non-linearity of the communication system.

20. The circuit of claim 15 , wherein when the double-end talk detecting device determines that double-end talk has occurred, the echo estimating adaptation process of the echo estimating coefficient is disabled.