A mathematical method used in signal processing, like audio compression, which operates on the principle that a sample of a signal can be predicted as a linear combination of past samples.