Method for detecting double-talk condition and system using the same
View Patent ↗A method for detecting a double-talk condition and an echo cancellation system using the same are introduced herein. According to an exemplary embodiment, the method may be adapted to an double-talk detector of an echo cancellation system and includes the following steps: obtaining an input signal with a first power; computing a second power, wherein the second power is a power combination of at least one estimated signal; determining whether the double-talk condition occurs according to a relationship between the first power and the second power, when the first power is larger than the second power, the double-talk condition occurs.
1. A method for detecting a double-talk condition, adapted to a double-talk detector of an echo cancellation system, the method comprising:
obtaining an input signal with a first power, wherein the obtaining the input signal with the first power comprises:
determining a first component of the input signal, wherein a frequency region of the first component is determined according to a linear region of an echo channel response; and
obtaining the first power by calculating a first component power of the first component;
computing a second power, wherein the second power is a power combination of at least one estimated signal, wherein the at least one estimated signal comprises an estimated echo signal and an estimated noise signal, wherein the computing the second power comprises:
determining a second component of the estimated echo signal and a third component of the estimated noise signal;
calculating a second component power of the second component; and
calculating a third component power of the third component, wherein the second power is computed by:
p
2
=
[
∑
k
∈
L
Y
~
(
k
)
2
+
∑
k
∈
L
N
~
(
k
)
2
]
+
η
∑
k
∈
L
N
~
(
k
)
2
,
where {tilde over (Y)}(k) and Ñ(k) respectively represents a k th frequency component of the second component and the third component, θ represents a margin factor, a frequency region of each of the second component and the third component is determined according to a linear region of an echo channel response, L represents the frequency region of the second component and the third component, wherein the margin factor is between 0 and a reciprocal of a signal-to-noise ratio of the input signal; and
determining whether the double-talk condition occurs according to a relationship between the first power and the second power, when the first power is larger than the second power, the double-talk condition occurs.
2. The method as claimed in claim 1 , wherein when the first power is smaller than the second power, determining that the double-talk condition does not occur.
3. The method as claimed in claim 1 , wherein the first component power of the first component is obtained by:
p
1
=
∑
k
∈
L
S
(
k
)
2
,
where S(k) represents a k th frequency component of the input signal, L represents a frequency region of the first component.
4. An echo cancellation system, comprising:
a microphone, configured to generate an input signal;
an echo canceller, coupled to the microphone and configured to generate an estimated echo signal and an error signal between the input signal and the estimated echo signal;
a noise processing unit, coupled to the echo canceller and configured to generate an estimated noise signal in response to the error signal;
a near-end voice detector, coupled to the microphone and configured to detect a near-end voice activity;
a far-end voice detector, coupled to the echo canceller, and configured to detect a far-end voice activity; and
a double-talk detector, coupled to the near-end voice detector, the far-end voice detector, the echo canceller and the noise processing unit, and configured to execute program instructions to execute a procedure for detecting a double-talk condition, the procedure comprising:
obtaining the input signal with a first power;
computing a second power, wherein the second power is a power combination of at least one estimated signal; and
determining whether the double-talk condition occurs according to a relationship between the first power and the second power, when the first power is larger than the second power, the double-talk condition occurs,
wherein the double-talk detector determines a first component of the input signal, and obtains the first power by calculating a first component power of the first component,
wherein the double-talk detector determines a frequency region of the first component according to a linear region of an echo channel response,
wherein the at least one estimated signal comprises the estimated echo signal and the estimated noise signal, the double-talk detector determines a second component of the estimated echo signal and a third component of the estimated noise signal where the double-talk detector determines a frequency region of each of the second component and the third component according to a linear region of an echo channel response, and respectively calculates a second component power of the first component and a third component power of the third component, wherein the double-talk detector computes the second power by:
p
2
=
[
∑
k
∈
L
Y
~
(
k
)
2
+
∑
k
∈
L
N
~
(
k
)
2
]
+
η
∑
k
∈
L
N
~
(
k
)
2
,
where {tilde over (Y)}(k) and Ñ(k) respectively represents a k th frequency component of the second component and the third component, η represents a margin factor, L represents the frequency region of the second component and the third component, the margin factor is between 0 and a reciprocal of a signal-to-noise ratio of the input signal.
5. The system as claimed in claim 4 , wherein when the first power is smaller than the second power, the double-talk detector determines that the double-talk condition does not occur.
6. The system as claimed in claim 5 , wherein the double-talk detector obtains the first component power of the first component by:
p
1
=
∑
k
∈
L
S
(
k
)
2
,
where S(k) represents a k th frequency component of the input signal, L represents a frequency region of the first component.
7. The system as claimed in claim 4 , wherein the echo canceller comprises:
an adaptive filter, coupled to the far-end voice detector and the double-talk detector, and configured to generate the estimated echo signal; and
an echo cancellation unit, coupled to the adaptive filter, the microphone and the noise processing unit, and is configured to generate the error signal by subtracting the estimated echo signal from the input signal.