Clustering of the Least Squares Lattice PARCOR (Partial Correlation) Coefficients: A Pattern-Recognition Approach to Steady State Synthetic Vowel Identification.

The partial correlation PARCOR coefficients of the least squares lattice filter may be used to conveniently and efficiently represent various types of acoustic signals. Because a stationary time series may be represented by a small number of PARCOR coefficients, the PARCOR coefficients have been widely used as effective pattern recognition parameters for the representation and transmission of information. This thesis establishes the PARCOR coefficients of the least squares lattice filter as efficient and effective pattern recognition features for the classification and identification of synthesized steady state vowel-like sounds. The PARCOR coefficient technique is shown to be a much quicker and more computationally efficient method of vowel identification than densification by formant frequencies, which involves the computations of poles and zeroes and the back-calculation of formant frequencies and format bandwidths. Author