Second formant effects of low-dimensional models of speaking rate change in articulatory synthesis

This study reports the initial developments of a method for simulating speaking rate change using articulatory synthesis. The method is driven by descriptive analyses of speech-kinematic and acoustic data. Emphasis is placed on representations of articulator speed, extent, and phase. The method can be used to explore hypotheses pertaining to normal and disordered speech production, articulatory-acoustic relations, and speech perception. The current work exploits the vocal tract modeling methods and software developed by Milenkovic (1998).; Speech kinematic and acoustic data were obtained from 5 speakers producing a wide range of speaking rates. Analyses of tongue movement speed suggested that speakers enacted complex, speaker-specific changes in tongue speed, particularly at extreme-slow rates of speech. Simple parametric descriptions of these complex changes show little consistency across speakers. Consequently, more coarse-grained summary-statistical descriptions of articulatory speed were employed to drive the rate simulations.; A preliminary method of rate simulation was devised and implemented based on low-dimensional representations of articulatory speed, extent, and phase. Initial simulations revealed plausible formant patterns, which micked certain characteristics of human speech. Similar patterns of change between articulatory and acoustic correlates were observed. Patterns of simulated changes in articulatory speed resulted in equivalent patterns in the rate of change of the 2 nd formant. Patterns of simulated changes in articulatory extent resulted in elaborated patterns of change in 2nd formant offset values. These simulated acoustic patterns appear to mimic the bi-modal distributions observed in rate-induced acoustic changes in human speech. Patterns of simulated changes in inter-articulatory phase revealed discontinuities in the resulting acoustic parameters. These discontinuities were traced to a methodological difficulty in simulating the very fine temporal resolution of the phase changes that occur in human speech. The results of this study have contributed to the identification of limitations and benefits of the current method of rate simulation. Moreover, it reports evidence inconsistent with the common finding that speed and extent of articulator movement are directly related.