Telephonic communications – Echo cancellation or suppression
Reexamination Certificate
1999-05-20
2001-07-31
Kuntz, Curtis (Department: 2643)
Telephonic communications
Echo cancellation or suppression
C379S413020
Reexamination Certificate
active
06269161
ABSTRACT:
TECHNICAL FIELD
The invention relates generally to voice communication systems that use a speakerphone to enable a hands-free voice communication. More specifically, the invention relates to acoustic echo cancellation techniques and full duplex speakerphone techniques that allow simultaneous reception and transmission of voice signals without significant switched loss or perceived echo.
DESCRIPTION OF THE RELATED ART
A full duplex speakerphone is a voice terminal that consists of at least one loudspeaker and at least one microphone, so that a hands-free voice communication is enabled. A full duplex speakerphone employs acoustic echo cancellation (AEC) techniques to permit simultaneous reception and transmission of speech without perceptible echo.
The fundamental structure of a typical full duplex speakerphone system is shown in FIG.
1
. The full duplex speakerphone (FDS) system
10
includes an attenuator
12
, a digital-to-analog (D/A) converter
14
, an amplifier
16
and a loudspeaker
18
that are coupled in series on a receive path
20
. The system also includes a microphone
22
, a second amplifier
24
, an analog-to-digital (A/D) converter
26
, a subtraction unit
28
and a second attenuation processor
30
that are coupled in series on a send path
32
. Situated between the receive path and the send path is an adaptive filter
34
. The adaptive filter and the subtraction unit define an acoustic echo canceller. The system further includes three measurement processors
36
,
38
and
40
that are each coupled to the receive or send path and an activity detection and control (ADAC) module
42
. The measurement processor
36
is coupled to the receive path. The measurement processor
40
is coupled to the send path between the A/D converter and the subtraction unit, while the measurement processor
38
is coupled to the send path between the subtraction unit and the attenuation processor
30
.
When an incoming digital signal from a far-end caller is received by the system
10
, the received signal is transmitted through the attenuator
12
on the receive path
20
. The received signal is then converted by the D/A converter
14
and amplified by the amplifier
16
. The amplified analog signal is broadcast into a room by the loudspeaker
18
. Depending on the acoustic characteristics of the room, an echo of the broadcast signal is propagated through various echo paths from the loudspeaker to the microphone
22
, such as echo paths
44
and
46
. The echo may be captured along with speech from the near-end caller by the microphone, and transmitted along the output path
32
as an outgoing analog signal. The outgoing signal is first amplified by the amplifier
24
and then converted into a digital format by the A/D converter
26
.
Meanwhile, the adaptive filter
34
samples the original received digital signal and performs a convolution step, i.e., a computation of an estimated echo response, using the sampled signal as a reference. The estimated echo response is a predicted acoustic echo response of the system when used in a particular environment. A current estimate of the echo response is utilized to subtract echo components of the outgoing signal. The subtraction, or cancellation, of echo components is performed by the subtraction unit
28
. After subtraction, this echo-cancelled outgoing signal is fed back to the adaptive filter
34
as an error signal that is utilized to dynamically adjust the filter coefficients that are used by the adaptive filter to execute the echo cancellation. The echo-cancelled outgoing signal is transmitted to the far-end caller via the attenuation processor
30
. Since there will usually be some residual echo, the attenuation processor is typically used to reduce the residual echo to an acceptable level. The attenuation processor can take the form of a device commonly known as the center clipper.
The quality of the echo cancellation depends on the ability of an adaptation algorithm, which is utilized by the adaptive filter
14
, to accurately model the true echo response. The true echo response is estimated by an adaptation process in which the error signal drives the adaptation algorithm to update the coefficients of the model, so that the error signal is driven toward zero, i.e., the echo-cancelled outgoing signal does not contain any detectable echo residue. However, the acoustic response of a room does not remain constant over time. For example, positional shifts of persons and/or the microphone in the room, or opening and closing of a door change the echo paths. The change in echo paths results in a new echo response that is inaccurately represented by the previously estimated echo model. Until the echo canceller can adapt to the new echo response, a substantial amount of echo may be transmitted to the far-end caller.
In addition to the echo cancellation, the system
10
is configured to determine if the signal activity is in one of four states: idle, far-end active, near-end active, or double-talk. If the far-end is active, the adaptive filter
34
should be allowed to further adapt, and the echo-cancelled outgoing signal may need to be suppressed by the attenuation processor
30
. However, if the echo-cancelled outgoing signal is mostly composed of valid near-end talker speech, then the activity state is in either the near-end active or double-talk state. In such state, the adaptive filter should be disabled and the send path attenuation should be set close to unity. This determination of the activity state is executed by the ADAC module
42
. The ADAC module examines the received signal, the original outgoing signal, and the echo-cancelled outgoing signal via the measurement processors
36
,
38
and
40
in order to determine the composition of the echo-cancelled outgoing signal.
In general, the conventional detection algorithm utilized by the ADAC module
42
to differentiate between the near-end talker speech and residual echo is based on ad hoc detection techniques. Most conventional implementations teach that a correlation measure is computed between the received signal and the error signal, i.e., the echo-cancelled outgoing signal. The correlation measure may be by time domain correlation or by frequency domain correlation. The correlation algorithm seeks to determine how much the error signal matches the received signal that was broadcast by the loudspeaker. If the error signal looks substantially similar to the received signal, the send signal is considered to be composed of mostly residual echo and only far-end activity is declared. However, if the error signal looks significantly different from the received signal, the send signal is considered to be composed of substantial near-end talker speech and a double-talk condition is declared.
In practice, it is quite difficult to apply a correlation measure to compare the error signal against the received signal. Typically, the acoustic echo lasts much longer than the observation window, and further, the transfer function between the loudspeaker and the microphone is very complex. In fact, the echo signal may look substantially different from the received signal. In addition to signal distortion caused by the acoustic response of the room, the correlation is further complicated by the fact that the adaptive filter
34
may be well converged at certain frequencies, but not as well at other frequencies. Furthermore, in some speakerphone systems even the transfer function of the loudspeaker is not known with any significant degree of certainty.
Notwithstanding these difficulties, the identification of the signal component of valid near-end talker speech within the echo-cancelled outgoing signal is relatively easy when the adaptive filter has perfectly cancelled the echo. However, in real situations, the acoustic echo path can change much more quickly than the adaptive filter
34
can re-adapt. Until the adaptive filter can re-adapt, a significant amount of residual echo will be introduced to the outgoing signal. If this residual echo is incorrectly identified as a valid near-end talk
King Linda J.
Logan Tamara L.
McLaughlin Hugh J.
Eng George
Kuntz Curtis
Law Offices of Terry McHugh
SignalWorks, Inc.
LandOfFree
System and method for near-end talker detection by spectrum... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for near-end talker detection by spectrum..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for near-end talker detection by spectrum... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2554096