Maximum likelihood methodology applied to empirically described probability distributions of speech movement data
H. Betty Kollia, Jay A. Jorgenson · The Journal of the Acoustical Society of America · 1999
In previous work [J. Acoust. Soc. Am. 102, 3164–3165(A) (1997)] an analysis of speech production data [from Kollia et al., J. Acoust. Soc. Am. 98, 1313–1327 (1995)] was initiated using Maximum Likelihood methodology in order to introduce a more realistic description of the behavior of the articulators. In this way the kinematic measures are described via natural probability distributions determined by the data set itself. These results show much smaller confidence intervals when the possible values of these parameters are described as lying in a bounded range (versus a normal distribution) with a diminishing likelihood of observing a data point farther from (than closer to) the mean. In this study the data are used to estimate the probability distributions (one for each value of the independent variable). The methodology employed, the family of probabilty distributions described by the data (one family for velocity-displacement data and another family for displacement-displacement data), and the resulting statistical analysis via maximum Likelihood methodology are reported on. Finally, the entire analysis is described in terms of what can be viewed as stochastic linear regression analysis with general error terms and the method for further analysis of error terms for comparable data. [Work supported by NSF.]