Databases Reference
In-Depth Information
Codebook index to channel
Input speech
Buffer
+
Excitat
ion
codebook
Variable
gain
Synthesized speech
−
Backward
gain
adaptation
50th-order
FIR filter
Backward
LPC
analysis
Perceptual
weighting
filter
Encoder
Codebook
index from
channel
Decoded speech
Excitation
codebook
Variable
gain
Adaptive
postfilter
Output
speech
Backward
gain
adaptation
50th-order
FIR filter
Backward
LPC
analysis
Decoder
F I GU R E 18 . 8
Encoder and decoder for the CCITT G.728 16kbps speech coder.
Just like the coders discussed previously, the sinusoidal coders divide the input speech into
frames and obtain the parameters of the speech separately for each frame. If we synthesized
the speech segment in each frame independent of the other frames, we would get synthetic
speech that is discontinuous at the frame boundaries. These discontinuities severely degrade
the quality of the synthetic speech. Therefore, the sinusoidal coders use different interpolation
algorithms to smooth the transition from one frame to another.
Transmitting all the separate frequencies
ω
l
would require significant transmission re-
sources, so the sinusoidal coders obtain a fundamental frequency
w
0
for which the approxi-
mation
K
(ω
0
)
A
y
n
=
ˆ
(
k
ω
0
)
cos
(
nk
ω
0
+
φ
k
)
(23)
k
=
1
