Analysis by Synthesis LPC Coding - Digital Speech: Coding for Low Bit Rate Communication Systems - page 195

7

Analysis by Synthesis
LPC Coding

7.1 Introduction

The broad classification of speech coding techniques that attempt to reproduce
the original speech waveform as best as possible can be split into two
basic groups, namely analysis-and-synthesis (AaS) schemes and analysis-by-
synthesis (AbS) schemes. Although AaS schemes, such as APC [1, 2], ATC [3]
and SBC [4], have been successful at rates around 16 kb/s and above, below
16 kb/s they can no longer reproduce good quality speech. In addition, AaS
coders that have been used at bit rates of around 9.6-16 kb/s can not achieve
true toll quality performance (MOS 4). There are two main reasons for their

shortcomings: first, the coded speech is not analysed to see if the coding
procedure is operating efficiently, i.e. there is no check on or control over the
distortions of the reconstructed speech; and secondly, in adaptive schemes,
the errors accumulated from previous frames are not usually considered in
the current frame of analysis, hence the errors propagate into the following
frames without any form of resetting. In AbS schemes, particularly AbS-LPC
schemes [5, 6], these two factors are incorporated in the coding process.
In AbS-LPC coding systems, a closed-loop optimization procedure is used
to determine the excitation signal, which produces a perceptually optimum
synthesized speech signal when used to excite the model filter .Itisthis
closed-loop approach which enables AbS-LPC coding schemes to be far more
successful at 4.8 to 16 kb/s than conventional AaS schemes such as APC and
SBC.
The method of AbS is not unique to speech coding, but is a general
technique used in other areas of estimation and identification. The basic idea
behind AbS is as follows. First it is assumed that the signal can be observed

≥