Digital Signal Processing Reference
In-Depth Information
x
(
n
)
FM radio signal
PT
S
LEM System
Model
TAP System
ASR
initialization,
SOU trigger
VAD Control
Notch Filter,
Noise Reduction,
VAD
h
(
n
)
v
( )
d
(
n
)
y
(
n
)
AEC,
Postfilter
s
(
n
)
Robust ASR
e
(
n
)
hands-free
microphone
n
(
n
)
Fig. 7.1 Block diagram of the talk-and-push (TAP) system
discrete-time domain, using
n
as discrete-time index at sampling frequency
f
s
¼
16 kHz,
the microphone signal can thus be expressed as the sum:
y
ð
n
Þ¼
s
ð
n
Þþ
d
ð
n
Þþ
n
ð
n
Þ
(7.1)
This relation is depicted on the bottom left of Fig.
7.1
.
To model the acoustic leaking from the loudspeaker into the microphone, we
assume that the echo signal
d
(
n
) results from the loudspeaker source signal
x
(
n
)by
convolution with a discrete-time, time-variant impulse response
T
h
ð
n
Þ¼
½
h
0
ð
n
Þ;
h
1
ð
n
Þ; ...;
h
N
1
ð
n
Þ
;
(7.2)
where
N
denotes the finite impulse response length and (·)
T
is the transpose.
For simplicity, a mono source signal
x
(
n
) is assumed. The impulse response h(
n
)
models the entire loudspeaker-enclosure-microphone (LEM) system—i. e., the
path from the digital-to-analog converter before the loudspeaker via the acoustic
enclosure to the analog-to-digital converter after the microphone.
Hence, the reverberated loudspeaker signal can be written as
h
T
d
n
ðÞ¼
ð
n
Þ
ð
n
Þ;
x
(7.3)
,
x
(
n
-
N
+ 1)]
T
is a
where
denotes the scalar product and
x
(
n
)
¼
[
x
(
n
),
x
(
n
-1),
...
time-inverted segment of the loudspeaker signal of length
N
.
As shown in Fig.
7.1
, the first stage of the TAP system is an acoustic echo
cancelation (AEC) unit. It computes an estimate
d
of the echo component
according to [
4
] and subtracts it from the microphone signal.
For this purpose, the LEM system transfer function is estimated using the FDAF
described in Sect.
7.3
. The FDAF furthermore contains a postfilter, which reduces
residual echo components as well as some background noise
n
(
n
) present in the
microphone signal.
ð
n
Þ
Search WWH ::
Custom Search