Joint maximum likelihood estimation of pitch and AR parameters using the EM algorithm

The speech production model where the speech signal is modeled as the output of an all pole filter driven either by some white noise sequence (unvoiced speech) or by the sum of an impulse sequence and a noise sequence (voiced speech) is considered. Approximate maximum-likelihood (ML) estimation algorithms for the unvoiced case are well known. In this work, the expectation-maximization (EM) algorithm is used in order to obtain the ML estimator of the parameters for the voiced speech model. These parameters consist of the parameters of the impulse sequence (pitch parameters) and the parameters of the filter (autoregressive parameters)