Linear Prediction
Wai C. Chu · 2003
Linear prediction (LP) forms an integral part of almost all modern day speech coding algorithms. The fundamental idea is that a speech sample can be approximated as a linear combination of past samples. In this chapter, the basic problem of LP analysis is stated, followed by its adaptation toward nonstationary signals. Examples of processing on actual speech samples are provided. Two computationally efficient procedures, namely, the Levinson–Durbin algorithm and the Leroux–Gueguen algorithm, are explained. The concept of long-term linear prediction is described, followed by some LP-based speech synthesis models. Practical issues related to speech processing are explained, with an alternative prediction scheme based on the moving average (MA) model given at the end of the chapter.