| Popis: |
Sets of pellet coordinates from the X‐ray Microbeam Speech Production Database, each corresponding to a static articulatory configuration, are submitted to a principal components analysis. Several talkers are analyzed separately, and various criteria are used to select the times in the continuous speech data from which to extract pellet data. In some instances, all of a talker’s speech data are submitted to an automatic procedure intended to select syllable nuclei, based on Mermelstein’s algorithm [P. Mermelstein, J. Acoust. Soc. Am. 58, 880–883 (1975)]. In other instances specific data are chosen by hand according to utterance type (e.g., vowel, glide, and consonant–vowel transition). These analyses are intended for use in a project to recover articulation from speech acoustics. In particular, we examine the relationship between pellet positions from data, the tongue shape reconstructed from the principal components analysis, and the parameters of a simple acoustic tube model, which is derived from the Stevens and House model [K. N. Stevens and A. S. House, J. Acoust. Soc. Am. 27, 484–493 (1955)]. [Work supported by Grant NIDCD‐01247 to Sensimetrics Corporation.] |
