This document provides an introduction and overview of linear predictive coding (LPC) and vocoding for speech processing.
LPC analyzes speech by estimating the spectral envelope (formants) and other parameters like intensity and pitch frequency. It removes the estimated formant effects to obtain the excitation signal. These parameters are transmitted instead of the full digital speech signal to reduce bandwidth.
Vocoding is a related speech analysis/synthesis technique. It uses a filter bank to extract the amplitude envelopes of different frequency bands, which are transmitted instead of the full speech signal. This also reduces bandwidth needed for voice transmission.
The document goes on to provide more details on the LPC analysis process, popular parameter representations like LAR