Digital Signal Processing Reference
In-Depth Information
fully unvoiced and the last being fully voiced. One method of deciding the
frequency marker is placing it at the end of the last voiced harmonic of
the spectrum, i.e. all the voiced harmonics are included in the voiced band
of the spectrum. A better solution for determining the frequency marker,
based on a soft decision process is described in [9]. The harmonic amplitudes
are estimated using equations (8.11) and (8.13) for voiced and unvoiced
harmonics respectively, however the LPC residual is used instead of the
speech signal. The LPC parameters are quantized and interpolated in the LSF
domain. The shape of the harmonic amplitudes is vector-quantized and the
gain is scalar-quantized separately.
At the receiving end, speech is synthesized with parameter interpolation
based on pitch cycle waveform (PCW). First, intermediate PCWs for the cur-
rent subframe are generated by interpolating the quantizedmodel parameters
of the last and current subframes. The excitation signal
e
i
(n)
,0
≤
n < T
0
,i
,for
the
i
th
PCW is produced as
V
c
e
i
(n
+
n
i
)
=
A
e,i
(l)
cos
{
lω
0
,i
(n
−
n
i
)
}
l
=
1
H
+
A
e,i
(l)
cos
{
lω
0
,i
(n
−
n
i
)
+
U
[
−
π , π
]
}
(8.25)
l
=
V
c
+
1
where H is the total number of harmonics,
ω
0
,i
=
2
π/T
0
,i
and
U
[
−
π , π
]
denotes a random number with uniform distribution between
−
π
and
π
.The
start position
n
i
for the
i
th
PCW is given by
i
−
1
n
i
=
n
0
+
T
0
,j
(8.26)
j
=
0
where
n
0
is the start position corresponding to the last position of the previous
subframe. The interpolated pitch
T
0
,i
for the
i
th
PCW is calculated as
α
i
T
(t
−
1
)
α
i
)T
(t)
T
0
,i
=
+
(
1
−
(8.27)
0
0
where
T
(t
0
is the received pitch of the
t
th
subframe. The interpolation factor
α
i
is defined as
G
(t)
N
i
α
i
=
(8.28)
G
(t
−
1
)
(N
−
N
i
)
+
G
(t)
N
i
where
N
is the subframe size,
G
(
·
)
is the received gain, and
N
i
is the PCW
position defined by,
0
.
25
(T
(t
−
1
)
T
(t
0
)
N
i
=
n
i
+
+
(8.29)
0
Search WWH ::
Custom Search