An adaptive speech analysis system

The process of speech production is approximated by an autoregressive moving-average process described by a state equation and an observation equation. The state transition matrix that characterizes the autoregressive part is determined to minimize the mean squared error in the estimation of the speech waveform by Kalman filtering. The Matrix and the autocorrelation function of the signal are used to determine the moving-average parameters. These parameters are then used to obtain the optimum number and the values of poles and zeroes of the speech power spectrum. Applications of the method to noise-stripping and pitch extraction are also discussed.