User Manual

Acoustic Echo Canceller White Paper
4. Specification
4.1. Measurement
We evaluated the acoustic echo canceller implemented
in chapter 3 in a conference room with a maximum
capacity of 180 people and a reverberation time of
approximately 1.0
s (Fig.5). In the teleconferencing
system, we used front
loudspeakers (the Yamaha
IF2108), ceiling
loudspeakers, and a microphone (the
Shure SM93). From the
loudspeakers, we output a signal
combining the signal received from the far end and the
microphone signal. We performed AEC processing on
the microphone signal and sent it to the far end. The
amount of acoustic echo is expressed as the difference
between the level of the input signal from the far end
and the level of the acoustic echo and is referred to as
ERL (echo return loss). The larger the ERL, the smaller
the acoustic echo and the more advantageous the
installation conditions. It is best to install a
teleconferencing system so that ERL is greater than 10
dB. We conducted this evaluation with the inherent ERL
at -1.2 dB, 5.5 dB, and 11.5 dB. An ERL value of -1.2
dB would constitute an extremely strong acoustic echo.
Fig.5 Setup
To evaluate echo suppression for single talk (in which only
the person speaking at the far end is talking), we used a CS
signal [4], and measured TCLw (weighted Terminal
Coupling Loss) . To evaluate conversation performance
for double talk, we used a CS signal and an AMFM signal
and measured the amount of echo suppression (TCL )
and the amount of insertion loss (A ) for the sound of
the near-end speaker .
[5]
[4]
wdt
H,S,DT
[6]
4.2. Result
Performance Evaluation for Single Talk
Fig.6 is the amount of echo suppression during single
talk (TCLw). ITU-T Rec. 341 recommends that echo
suppression be not less than 35 dB, but regardless of the
amount of ERL, a high amount of echo suppression,
approximately 70 dB, is achieved. It is clear that the
acoustic echo canceller can reliably remove acoustic
echo.
[7]
Fig.6 TCLw
Performance Evaluation for Double Talk
Fig.7 is the amount of echo suppression during double
talk (TCL
wdt
). During double talk, because both speakers
are talking at the same time, acoustic echo stands out
less than it does with single talk. The necessary amount
of echo suppression varies depending on the volume of
the sound being produced by the loudspeakers, the
sensitivity of the microphone, and other settings, but
given the measurement conditions of this evaluation, it
is presumed that echo suppression of not less than 15 to
25 dB is necessary (ITU-T Rec. P.340
[8]). When the
amount of acoustic echo is low (ERL = 11.5 dB),
sufficient echo suppression is achieved even when AEC
is set to type 1 (softest). When the amount of acoustic
echo is high (ERL = -1.2 dB), it is necessary to set AEC
to 2 or higher to suppress the acoustic echo.
Fig.8 shows the amount of insertion loss (A
H,S,DT
) for
the sound of the near-end speaker during double talk.
When the insertion loss is great, the possibility that the
sound will be cut off or that the quality of the sound will
change increases. According to ITU-T Rec. P.340
[8], an
insertion loss of not more than 3 dB is recommended.
Even when the amount of acoustic echo is high (ERL=
3