Telephonic communications – Echo cancellation or suppression – Using digital signal processing
Reexamination Certificate
2001-01-26
2004-09-14
Barnie, Rexford (Department: 2643)
Telephonic communications
Echo cancellation or suppression
Using digital signal processing
C379S406120, C379S406010, C379S406130, C379S390020, C370S286000, C370S290000
Reexamination Certificate
active
06792107
ABSTRACT:
TECHNICAL FIELD
This invention relates to echo cancelling in general and to double-talk detection in particular.
BACKGROUND OF THE INVENTION
Double-talk occurs when both parties to a telephone call are talking at the same time. An acoustic echo canceller (AEC), employed to cancel echoes at either end of the call, needs to be able to detect double-talk in order to operate properly. An AEC identifies the impulse response of the acoustic echo path between the near-end (local) loudspeaker and microphone, generates a replica of the echo by using an adaptive filter, and subtracts the echo replica from the microphone output to cancel the echo of the far-end speech. The adaptive filter uses the far-end speech as a reference signal. If the adaptive filter is allowed to adapt in the presence of near-end speech, the near-end speech will be added to the error signal which drives filter tap coefficient adaptation, thereby corrupting the estimate of the acoustic echo path. Therefore, the acoustic echo canceller must suspend its adaptation to the impulse response of the acoustic echo path while double-talk is present.
The prior art includes two approaches to double-talk detection. One approach assumes that the energy level of the echo is a given fraction of the energy level of the near-end speech. But this is an invalid assumption in the case of a telephony-enabled personal computer (PC), where arbitrary gain can be applied to the loudspeaker independently of the PC, and hence the echo can actually be greater than the original reference signal and even greater than the near-end speech. The other approach is to examine the content of the signal by using cepstral analysis, thereby allowing individual talkers to be identified within the signal. But this is a highly-complex solution, too computationally-complex and slow to be carried out during real-time operation on a typical end-user PC. Hence, the prior art does not offer a suitable technique for double-talk detection in a telephony-enabled PC.
SUMMARY OF THE INVENTION
This invention is directed to solving these and other problems and disadvantages of the prior art. According to the invention, double-talk detection is effected as follows. A cross-correlation is effected between a received far-end signal and a transmitted near-end signal to determine their level of convergence, signal energy of the transmitted near-end signal is determined, and average signal energy is determined from the signal energy of the transmitted near-end signal and signal energy of any previously-transmitted near-end signal. Preferably, prior to the above activities the transmitted near-end signal and the received far-end signal are transformed to make double-talk detection more efficient, such as by being converted to the frequency domain. Also preferably, prior to those above activities the near-end and far-end signals are filtered to remove noise signal energy therefrom. Following the above activities, double-talk detection is enabled if a comparison of the level of convergence with a first threshold indicates that the signals are converged, and double-talk detection is disabled if a comparison of the level of convergence with a second threshold indicates that the signals are diverged. If double-talk detection is not enabled, then an error threshold is set to the average signal energy. If double-talk detection is enabled, presence of double-talk is indicated if the signal energy of the transmitted near-end signal exceeds the error threshold, and absence of double-talk is indicated if the signal energy of the transmitted near-end signal does not exceed the error threshold.
Advantageously, the invention is not dependent upon any preset energy thresholds (i.e., it does not require assumptions and predictions about relative energy levels to be made), and it is computationally efficient.
While the invention has been characterized in terms of a method, it also encompasses apparatus that performs the method. The apparatus preferably includes an effector—any entity that effects the corresponding step, unlike a means—for each step. The invention further encompasses any computer-readable medium containing instructions which, when executed in a computer, cause the computer to perform the method steps.
These and other features and advantages of the invention will become more apparent from the following description of an illustrative embodiment of the invention considered together with the drawing.
REFERENCES:
patent: 5463618 (1995-10-01), Furukawa et al.
patent: 5696821 (1997-12-01), Urbanski
patent: 5721772 (1998-02-01), Haneda et al.
patent: 5732134 (1998-03-01), Sih
patent: 5761318 (1998-06-01), Shimauchi et al.
patent: 6049606 (2000-04-01), Ding et al.
patent: 6185300 (2001-02-01), Romesburg
patent: 6192126 (2001-02-01), Koski
patent: 6563803 (2003-05-01), Lee
Tucker Luke A.
Wildie Mark Greig
Barnie Rexford
Lucent Technologies - Inc.
LandOfFree
Double-talk detector suitable for a telephone-enabled PC 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 suitable for a telephone-enabled PC, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Double-talk detector suitable for a telephone-enabled PC will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3251677