Telephonic communications – Subscriber line or transmission line interface – Network interface device
Reexamination Certificate
1998-05-28
2001-02-20
Woo, Stella (Department: 2743)
Telephonic communications
Subscriber line or transmission line interface
Network interface device
C379S406010, C379S387020
Reexamination Certificate
active
06192126
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a method for double talk detection and a double talk detector, especially adapted to be used with or in an echo canceller, for detecting double talk situations in a device capable of two-way voice communication, the double talk detector comprising means for receiving a speech signal, means for dividing the speech signal into subsignals representing specific frequency bands, means for performing a detection for each subsignal, and means for calculating subdecision signals on the basis of the detection, the subdecision signal indicating existence of double talk in the respective subsignals. The invention also relates to a device incorporating such a double talk detector.
BACKGROUND OF THE INVENTION
Telephone communication is often disturbed by echo. This especially applies to full-duplex telephones which have four possible modes of operation: idle, near-speech, far-speech, and double-talk. Echo usually occurs in far-speech situations, when the received far-end signal reproduced by the speaker is caught by the microphone and thereby returns to far-end. A full-duplex telephone requires adaptive signal processing techniques to cancel acoustic feedback or echo. A known approach to avoid disturbing echo is to use an echo canceller or echo suppressor. Echo cancellers which are well known in the telephone communications environment usually employ a form of an adaptive digital filter. The echo canceller typically functions to disable the outgoing path from a phone when a signal from the far-end (speech received by the phone) appears on the incoming path. Therefore, echoes due to incoming signals on the receive path are prevented from returning to far-end over the outgoing path. Echo cancellation is usually implemented so that the parameters of the adaptive filter performing the echo cancellation are updated whenever far-end speech occurs in order to achieve echo cancellation as accurate as possible for each situation.
Double talking refers to the condition when the near end subscriber (the user of the phone) and the far-end subscriber talk simultaneously. When both parties talk simultaneously, i.e. during double talk, the echo canceller is no longer able to effectively block echo signals. This is because the echo signals are included in the near-end subscriber's signals to be transmitted, i.e. a desired signal to be transmitted and an echo signal are simultaneously applied to the send input. The super-positioning of these signals causes distortion of the adjustment of the echo canceller when it considers both the echo signal and the desired signal to be transmitted. This means that the replica produced by the echo canceller no longer sufficiently cancels the current echo signal. Accordingly, it is a current practice to provide a double talk detector for preventing the disturbing influence of double talk on echo canceller adjustment. This means that the parameters of the adaptive filter performing the echo cancellation are not updated during double talk. Echo and double talk are problems especially in speaker phones and in phones with hands-free equipment in which the far-end signal from the speaker is captured by the microphone.
Prior art echo suppressors include double talk detectors which distinguish between near-end speech, i.e. speech signals generated on the outgoing path by the near-end subscriber, and echo signals returning on the outgoing path due to far-end subscriber speech signals on the incoming path. If the outgoing path signal exceeds the incoming path signal it is assumed that the near-end subscriber is transmitting and the echo suppression is disabled. When the opposite condition occurs it is assumed that the near-end subscriber is not transmitting and the echo is suppressed.
FIG. 1
shows a block diagram of a phone comprising an echo canceller
1
known from prior art. The near-end signal
3
comes from the microphone
2
and is detected by a near-end voice activity detector
4
(VAD, Voice Activity Detector).
The far-end signal
5
comes from an input connection I of the phone (e.g. wire connection in wire phones and from air interface/antenna to reception branch in a mobile phone) and is detected by a far-end voice activity detector
6
and is finally output by the speaker
7
. Both near-end signal
3
and far-end signal
5
are fed to a double talk detector
8
for double talk detection and to an adaptive filter
9
for performing echo cancelling. The adaptive filter
9
also receives the output of the double talk detector
8
in order to avoid adaptation of the filter during double talk. The adaptive filter
9
outputs a signal
10
which is subtracted from the near-end signal
3
in a summing/subtracting means
11
for cancelling echo and getting an echo cancelled output signal
12
, which is forwarded to output connection
0
(e.g. wire connection in wire phones and from transmission branch to air interface/antenna in a mobile phone). The operation principle used is that the double-talk detector
8
needs the information of far-end speaker active and so it works only when there is far-end speech
5
and it is quite easy to detect if there is only far-end
5
or only near-end
3
speech. The adaptive filter
9
is controlled by double-talk, near-end and far-end detectors
8
,
4
and
6
. If there is double-talk, the coefficients of the adaptive filter
9
are frozen. The device of
FIG. 1
could as well be a hands-free equipment with speaker, microphone, input and output connectors to a telephone device and the echo canceller situated in the hands-free device.
In European patent publication EP-B1-0 053 202 double talk detection is based on three detectors. A first detector compares the amplitude of the near-end signal before adaptation with the amplitude of the near-end signal after adaptation resulting in a first amplitude ratio. A second detector compares the amplitude of the near-end signal with the amplitude of the far-end signal resulting in a second amplitude ratio and a third detector compares the amplitude of the near-end signal before adaptation with the amplitude of the far-end signal resulting in a third amplitude ratio. The outputs of each detector are combined to make a common decision. The outputs are the above mentioned amplitude ratios taken over the entire frequency band. A drawback of this solution is that the result of adaptation affects the double talk detection so that actually a reliable detection is achieved only when the adaptation is false.
European patent publication EP-A2-0 439 139 discloses a double talk detector in which detection is based both on a cross-correlation value between the incoming signal and the echo signal and on a power ratio between the same signals. The power ratio is calculated over the entire frequency band. A drawback of this solution is that the changes of the cross-correlation value are slow, which makes double talk detection slow. The cross-correlation value is an average over several speech frames. The faster the double talk detection is the faster the adaptation (of the filter performing the echo suppression) can be stopped and false adaptation can be avoided. Likewise, when double talk detection is slow more false adaptation occurs.
European patent publication EP-A1-0 454 242 discloses an echo canceller in which double talk detection is performed by dividing the frequency band into narrower bands, i.e. into sub-channels. Then double talk detection is done separately for each sub-channel on the basis of the power ratio of a receive input signal and a send output signal. The detection result of each sub-channel is used only for adjustment of the adaptation of the same sub-channel in question. A drawback of this solution is that because the results are interpreted separately, a false echo cancellation result may be achieved if all sub-channels having double talk are not detected. Even if adaptation is stopped for a sub-channel in which double talk was detected a possibility exists that in a nearby sub-channel double talk exists as well, for which chann
Nokia Mobile Phones Ltd.
Perman & Green LLP
Woo Stella
LandOfFree
Double talk detector, method for double talk detection and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Double talk detector, method for double talk detection and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Double talk detector, method for double talk detection and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2609465