Estimation of the parameters of an autoregressive process in the presence of additive white noise

Applications of linear prediction (LP) algorithms have been successful in modeling various physical processes. In the area of speech analysis this has resulted in the development of LP vocoders, devices used in digital speech communication systems. The LP algorithms used in speech and other areas are based on all-pole models for the signal being considered. With white noise excitation to the model, the all-pole LP model is equivalent to the autoregressive (AR) model. With the success of this model for speech well established, the application of LP algorithms in noisy environments is being considered. Existing LP algorithms perform poorly in these conditions. Additive white noise severely effects the intelligibility and quality of speech after analysis by an LP vocoder.