Analysis by Synthesis LPC Coding - Digital Speech: Coding for Low Bit Rate Communication Systems - page 197

Original
Speech

Encoder

s(n)

^ +

Pitch
Synthesis
Filter

LPC
Synthesis
Filter

s(n)

Excitation
Generator

−

+

Error
e(n)

Error
Minimization

Decoder

Output
Speech

Pitch
Synthesis
Filter

LPC
Synthesis
Filter

Optimum
Excitation

^

s(n)

Figure 7.2 Block diagram of AbS-LPC coding scheme

model requires frequent updating of the parameters to yield a good match
with the original signal, the analysis procedure of the system is carried out
in blocks, i.e. the input speech is partitioned into suitable blocks of samples.
The update rate of the analysis block or frame determines the bit rate or
capacity of the coding schemes. The basic operation of an AbS-LPC scheme
is as follows:

1. Initialize the contents of the time-varying filter (LPC and pitch) to pre-
determined values (usually zero or low level random noise).
2. A frame of speech samples is buffered and a set of LPC coefficients are
computed, using LPC analysis on the frame.
3. As the LPC analysis frame is usually too large for efficient analysis to
determine the excitation, the frame is subdivided into a subframes.
4. For each subframe:
(a) Using the computed LPC coefficients (usually interpolated for each
subframe) in the LPC filter, its memory effect (filter-ringing response)
is computed and subtracted from the original signal, which is usually
perceptually weighted.
(b) The pitch filter delay (the pitch or its integer multiples) and its associ-
ated scaling factor (pitch gain) are then calculated. This calculation is
performed in such a way that the difference between the synthetically-
generated speech and the remaining original signal is minimized.
(c) Once the pitch filter parameters are found, the pitch and LPC synthesis
filter can be grouped together to form a cascaded filter. Using this
cascaded filter, the best secondary excitation is determined in such a