Effects of consonant context on vowel formant contours in spontaneous and read speech

Abstract

A database of recordings from 163 speakers from Nova Scotia, Canada was collected with the aim of comparing formant contours between spontaneous and read speech. In the reading task, participants were asked to produce a number of real and nonsense words spanning the inventory of vowels in this dialect in a variety of consonant contexts. In the second part of the reading task, speakers were asked to read 20 sentences from the TIMIT database. These latter recordings were used to assist in the training of the force-alignment system which used to segment the recordings into phonemes from a text transcript. In addition to the reading task, speakers also provided a monologue on a topic of their choice. These recordings were screened for disfluencies, noise, and disruptions, manually segmented into breathgroups, and then transcribed. Stressed vowels in /CVC/ contexts, where C corresponds to plosives, were sampled from both the read and spontaneous speech. Formants were tracked and measured automatically, and an analysis similar to that of Broad and Clermont [J. Acoust. Soc. Am., 81, 155–165 1987] was performed in which consonant effects on formant transitions are treated as additive effects. [Work supported by SSHRC.]