Digital Signal Processing Reference
In-Depth Information
, the speech and silent passages within
this signal can be detected. Therefore, the squared magnitude of
s
By using the predefined test signal
s
ð
n
Þ
ð
n
Þ
is calculated
and smoothed over the time:
2
2
2
s
ð
n
Þ
¼ a j
s
ð
n
Þj
þð
aÞj
s
ð
n
Þj
:
j
j
1
1
(5.1)
. By means of this smoothed
discrete signal, the set of the sampling points for the speech and pause passages can
be derived by
The smoothing factor is chosen as
a 2
½
0
:
001
;
0
:
01
n
o
2
T
speech
¼
n
j
s
ð
n
Þ
j
>
S
0
;
(5.2)
n
o
2
T
pause
¼
n
j
s
ð
n
Þ
j
<
N
0
;
(5.3)
where
S
0
gives the threshold for the speech passage detection and
N
0
for the pause
passages; in addition,
S
0
N
0
must hold. It is assumed that the signal recorded by
the binaural microphones is defined as
y
i
ð
n
Þ¼
h
i
ð
n
Þ
s
ð
n
Þþ
b
ð
n
Þ¼
u
i
ð
n
Þþ
b
ð
n
Þ;
(5.4)
where
b
the impulse response between the
artificial mouth and the corresponding
i
-th binaural microphone, and
u
i
ð
ð
n
Þ
indicates the additive noise,
h
i
ð
n
Þ
is equal to
the convolution of the impulse response and the test signal. By combining these
definitions, the noise power
P
B;i
and the noisy speech power
P
Y;i
can be estimated by
n
Þ
X
n 2 T
pause
j
1
2
P
B;i
¼
T
pause
y
i
ð
n
Þj
(5.5)
#
and
0
1
X
1
@
2
A
P
U;i
¼
T
speech
j
y
i
ð
n
Þj
P
B;i
;
(5.6)
#
n 2 T
speech
where
defines the cardinality of a given set. Hence, the logarithmic SNR in
dependence of the ear is defined as
#
P
U;i
P
B;i
P
B;i
SNR
i
¼
10
log
10
P
U
;
i
P
B;i
¼
10
log
10
1
:
(5.7)
Search WWH ::
Custom Search