Modelling of vowel-inherent spectral change in spontaneous and elicited speech

Abstract

A database of speech samples was collected to examining vowel formant patterns in spontaneous and read speech. Recordings were made from 220 long-time residents of Nova Scotia and Prince Edward Island in two separate tasks: Participants read a series of sentences in several phonetic contexts and also provided a monologue of at least ten minutes duration resulting in approximately 36 hours of spontaneous speech. Vowel formant frequencies were measured for all stressed vowels in non-function words surrounded by obstruents. Speakers were divided among eight geographic regions. Consonant contexts were modelled via nonlinear regression based on a model inspired by Broad and Clermont (J. Acoust. Soc. Am. 81,155). Spontaneous speech was characterized by retraction of front vowels, fronting of /u/, backing of /æ/, and greater diphthongization of /ɒ/. Older speakers were less likely to show a merger between /ɒ/ and /ɔ/. There was also some evidence of a partial Canadian Shift whereby /æ/ and /ε/ are lowered following merger of /ɒ/ and /ɔ/. Results from alternate models currently under development, which may better accommodate vowel-inherent formant movement, will be discussed. [Work supported by SSHRC.]