Digital Signal Processing Reference
In-Depth Information
25.0
5 ms, with LTP
10 ms, with LTP
5 ms, no LTP
10 ms, no LTP
20.0
15.0
10.0
5.0
0.0
0.0
10.0
Decimation factor
20.0
Figure 7.11
SNR plot of RPELPC with different configurations
performance at a lower analysis frame size is better. This is because it has
more degrees of freedom to vary the excitation to match the reference vector.
7.3 Code-Excited Linear Predictive Coding
Amongst the variations of AbS-LPC schemes, the most widely-reported
scheme used at 8 kb/s and below is code-excited linear predictive coding
(CELP). As the name suggests, the excitation of the time-varying filters is
provided by a codebook. CELP operates as follows (see the simplified block
diagram in Figure 7.12):
1. The original speech,
s(n)
, is partitioned into analysis frames of around
20-30ms. LPC analysis is performed on the frame of
s(n)
to give a set of
LPC coefficients which are used in the short-term LPC predictor to model
the spectral envelope of the speech.
2. Having computed the LPC parameter, the frame is usually split into a
number of subframes (usually 40 or 60 samples long). The following
processing is carried out for each subframe.
Search WWH ::
Custom Search