Linear Predictive Coding
If one approximates the vocal tract as a series of fixed length tubes (which is equivalent to representing it as an all-pole digital filter) it becomes possible to predict successive samples of the speech wave as linear combinations of previous samples. The coefficients in the linear combination characterize the shape of the vocal tract. A sequence of sets of coefficients can be used to characterize the changing shape of the vocal tract over time. This representation is widely used because of the particularly efficient algorithms associated with it.