MPEG-4 Book, The

10.1. Introduction to Speech Coding

In contrast to general audio coders for natural audio signals, speech coders are usually expected to handle human voice as an input signal. This means that a model-based signal analysis and synthesis can be used in a highly efficient manner. The most common and widely used model-based speech analysis system is linear predictive (LP) analysis, also known as linear predictive coding (LPC). In LPC analysis, a speech signal is decomposed into two components: a set of LP coefficients and a prediction error signal (residue). The LP coefficients allow the construction of an analysis filter, which removes short-term correlation in the speech signal and provides a prediction error signal. In this analysis, a speech ...