Despite its usefulness in reducing the bit rate for digital speech transmission, the production of natural-sounding speech using conventional LPC techniques has proved difficult. This is largely due to the simplistic model of the excitation source (pitch pulses or white noise). Attempts to overcome the deficiencies in this binary model include residually excited LPC and multi-pulse modelling of the excitation source. However these approaches require a substantial increase in the bit rate over conventional LPC. An alternative approach is code excited LPC which can achieve bit rates as low as 4.8kb/s but at a cost of great computational effort making real time implementation unfeasible.